That’s the term for when AI sounds completely sure of itself, and it’s just wrong.

Here’s the whole thing, in one lie.

Ask an AI who invented the light bulb, and it might tell you: 1879, Dr. Alaric Brenton.

No hedging. No “I think.” Just a name, a year, delivered like a fact you’d find on a plaque.

There’s one problem. Dr. Alaric Brenton doesn’t exist. The light bulb was Edison’s, and every schoolkid knows it. The AI didn’t know it was lying to you. It didn’t even know it was guessing.

That’s a hallucination. A confident, fluent, completely wrong answer, dressed up exactly like a true one. You can’t patch it out like a bug, because it isn’t one.

Why it happens

AI is trained on an enormous pile of everything: books, code, conversations, blog posts, images, graphs, whatever it could get its hands on.

From that pile, it learns patterns. What word tends to follow what word. What an answer to a question like yours usually looks like.

So when you ask it something, it’s not looking anything up. It’s predicting the most probable next words, based on the patterns it’s seen before. Most of the time, that prediction lands close to true. Sometimes it doesn’t, and you get Dr. Alaric Brenton.

The AI isn’t choosing to deceive you. It’s doing the same thing it always does, guessing the next likely word, and this time the guess missed.

How to actually use this

Use AI to help you. Don’t use it to replace your judgment.

Ask it for a first draft, a summary, a starting point. Then check the facts. Ask it for sources. Look them up yourself before you repeat them to anyone.

Think of AI as your intern: fast, tireless, genuinely useful, and not the one who signs off on the final answer. That’s still you.

That’s a wrap on AI hallucinations. See you tomorrow with another one.

Got a complex AI topic you want broken down like this? Drop it in the comments, and I’ll cover it in an upcoming article.

I know. The word alone makes people’s eyes glaze over. Application Programming Interface. Sounds like something only engineers are allowed to understand.

It’s not. You already understand it. You just don’t know you do yet.

Here’s the whole thing, in one scene.

You walk into a restaurant. You tell the waiter you want a burger. The waiter doesn’t cook it. The waiter walks your order to the kitchen, hands it to the chef, waits, and brings the finished burger back to your table.

You never talked to the chef. You didn’t need to. The waiter handled the whole exchange.

That waiter is an API.

Now put that in your phone.

Say you open your weather app to check Chennai before you leave the house. The app doesn’t know the weather. It has no idea what’s happening in the sky over Chennai right now.

So it sends a request, through an API, to a weather service, maybe Google Weather, maybe some other provider working quietly in the background. That service checks its data, sends the answer back through the same API, and your app displays “32°C, humid, don’t bother with the jacket.”

You saw a temperature. What actually happened is your app asked a stranger a question and got a straight answer back, fast enough that it felt instant.

One app, many waiters.

This is where it gets useful to understand, not just cute.

Open Swiggy or Uber Eats to order dinner. That single app is running several of these API conversations at once. One API checks your delivery address against maps. Another one processes your payment. Another confirms your login. If something goes wrong and you open a support chat, that’s routed through an API too, sometimes straight to an AI system on the other end.

None of these systems talk to each other directly. They all go through the API, get their answer, and hand it back to the app you’re staring at. You just see one clean screen. Behind it, a small army of waiters is running back and forth to a small army of kitchens.

That’s the whole concept. A messenger that lets one piece of software ask another piece of software a question and bring back the answer, without either one needing to know how the other actually works.

You don’t need to know how the kitchen makes the burger. You just need someone to bring it to your table.

Got a topic you want explained without the nerd stuff? Drop it in the comments and I’ll cover it. I’m teaching a million people AI by the end of 2026. If you’re reading this in July, you’ve got until December to become genuinely good at this.

See you tomorrow.

Get the full AI Vocabulary

AI for Non-IT Library | The Arthi AI Collective

You would’ve probably used a Chatbot and called it an AI Agent.

Let’s learn AI without the nerd stuff. Day 2. Today, we talk about AI agents.

To get this, you first need to see the gap between a chatbot and an agent. Because right now, most people don’t.

Chatbots give you ideas. Agents do the work.

ChatGPT, Gemini, Claude, all the tools you already use for a quick prompt: chatbots.

Ask one to plan a farewell party and you’ll get a list. Cake ideas. A playlist. A few party games. Useful, but it stops there. You still have to do everything on that list yourself.

Give an AI agent the same prompt, “plan a farewell party,” and it doesn’t stop at ideas. It sets a budget. Builds a guest list. Checks a couple of venues. Sends the invites. Sets reminders. Builds you a checklist.

That’s the whole difference in one sentence: a chatbot answers you, an agent works toward a goal for you.

How an agent actually gets there

Say you ask an agent to help you study for a science test. Here’s the sequence:

1. Goal. Study for the test. That’s the target everything else points at.

2. Plan. It builds a real study schedule, not a vague outline.

3. Tools. It searches the web for resources, checks your calendar for free time, pulls files you already have.

4. Action. It drafts the study plan, blocks the time, puts it in front of you.

5. Review. You check it. If something’s off, it goes back and adjusts.

That loop, plan, act, review, repeat, is what makes it an agent instead of a one-shot answer.

You are still in charge. Always.

This is the part people skip, and it’s the one that matters most.

An agent can save you hours. It cannot make the call for you. You give it the goal. It does the busy work: comparing options, researching, drafting, scheduling. Think of it as a genuinely capable digital intern.

But an intern doesn’t send the client email without you reading it first. Same rule here.

Use an agent to draft a message, organize your files, or run research. Let it save you the grunt work every time.

But if it’s a financial decision, or a message going out under your name, you review it before it goes live. Every time. No exceptions.

That’s not a limitation. That’s the whole point of the system working correctly.

That’s a wrap on AI agents. Tomorrow, another AI topic, still no nerd stuff.

Got a topic you want broken down like this? Drop it in the comments, and follow along for the rest of the series.

Get the full vocabulary here👇

But here’s the truth: you have probably already used Artificial Intelligence today.

When Netflix recommended a show, when Instagram guessed which Reel you might like, when Google Photos found your old beach pictures, or when ChatGPT helped you write something, AI was quietly working in the background.

Today, let’s make it simple.

Why Artificial Intelligence Matters

Artificial Intelligence matters because it is becoming part of almost every tool we use.

You do not need to become a computer scientist to understand AI.

But if you understand the basic idea, you will stop feeling scared of it and start using it better.

You’ve already used AI. You just didn’t know its name.

Artificial Intelligence in One Line

Artificial Intelligence is technology that learns patterns from data and uses those patterns to make predictions, create things, or help solve problems.

That’s it.

AI is basically a very powerful pattern-finding machine.

An Everyday Analogy

Imagine a 15-year-old student who has watched thousands of cooking videos.

They have seen how people make pizza, pasta, biryani, noodles, cakes, sandwiches, and salads.

Now, one day, you ask them:

“Can you suggest a quick dinner recipe using bread, cheese, and vegetables?”

They may not have seen that exact recipe before.

But because they have seen many similar recipes, they can guess what might work.

They may say:

“Toast the bread, add cheese, add vegetables, season it, and grill it.”

That is not magic.

That is pattern learning.

AI works in a similar way.

It looks at a huge number of examples, learns patterns from them, and then uses those patterns to answer questions, generate images, recommend videos, translate languages, or help with tasks.

It is like a student who has studied millions of examples, but instead of studying in a classroom, it studies data.

What Is Actually Happening in AI?

Let’s say you ask ChatGPT:

“Explain photosynthesis like I’m 10.”

ChatGPT does not have a tiny teacher sitting inside it.

It has learned from a huge amount of text. It has seen how people explain science, how teachers simplify ideas, how children’s books describe plants, and how examples are usually structured.

So when you ask your question, it predicts a useful answer based on patterns it has learned.

A simple version of what happens is this:

AI studies examples.
It finds patterns.
It uses those patterns to make predictions.
Then it gives you an output.

That output can be a sentence, an image, a voice, a video, a recommendation, a summary, or even a decision.

For example, if you type:

“The sun rises in the…”

Most people will complete it as:

“east.”

AI also does something like this, but at a much bigger level. It predicts what should come next based on patterns.

Modern AI can work with words, images, audio, video, numbers, and even scientific data.

That is why AI is not just one tool. It is a whole family of technologies.

Where You See AI Daily

Common Misconception

A lot of people think AI “understands” everything like a human.

It does not.

AI does not have feelings, life experience, wisdom, or common sense like we do.

It can sound confident even when it is wrong. That is why you should not blindly trust everything AI says.

Think of AI like a very fast assistant.

It can help you think, write, search, organize, and create.

But you are still the human in charge.

Recap

Artificial Intelligence is technology that learns from data.

It finds patterns and uses them to make predictions.

AI powers tools like ChatGPT, Netflix, Instagram, Google Photos, Amazon, and Google Maps.

AI can create text, images, audio, video, summaries, recommendations, and more.

AI is powerful, but it is not perfect. You still need human judgment.

Comment one AI concept you’ve always pretended to understand. I’ll decode it next, without the computer-science headache.

I need you to sit down for this one.

Because March 2026 wasn’t just a big month for AI. It was the kind of month that makes you look back a year from now and go: “Oh. That’s when everything changed.”

Five frontier models launched in 23 days. OpenAI killed Sora, its flagship video product, after burning $15 million a day on it. Anthropic sued the Pentagon and won (for now). xAI got hit with lawsuits on three continents. And $10 billion in funding closed in a single week.

I made you a 12-slide carousel, breaking it all down. But if you want the full story with all the context? You’re already reading it.

Have you subscribed to my YouTube?

Let’s go.

This Month in One Line

Every major AI lab shipped a new model, OpenAI deleted one of theirs, Anthropic took the U.S. government to court and somehow the robots also had a moment.

The Big Story: Five Frontier Models in 23 Days

We have never seen anything like this.

In the span of three weeks, every major AI lab shipped a new frontier model. Not incremental updates. Not minor patch releases. Full frontier upgrades, all within a 23-day window.

Here’s your cheat sheet:

• GPT-5.4 (OpenAI, Mar 5) : 1.05 million token context window. Three variants: Standard, Thinking, and Pro. Scored 83% on GDPval, a benchmark designed to measure economically valuable work. Pricing starts at $2.50 per million tokens.

• Gemini 3.1 Pro (Google, Mar 11) : Tied for #1 on the Artificial Analysis Intelligence Index. 77.1% on ARC-AGI-2. Matched GPT-5.4 Pro at the top of the leaderboard.

• Grok 4.20 Beta (xAI, Mar 3) : The weird one. A four-agent collaboration system where Grok coordinates, Harper researches, Benjamin handles logic, and Lucas plays devil’s advocate. A 2-million-token context window. API pricing dropped 60% from Grok 3.

• Qwen 3.5 Small (Alibaba, Mar 1) : The biggest surprise. A 9-billion-parameter open-source model that scored 81.7% on GPQA Diamond, outperforming OpenAI’s own 120-billion-parameter open model. The 2B variant runs on an iPhone in airplane mode. Under Apache 2.0 license.

• Nemotron 3 Super (NVIDIA, Mar 11) : The highest-scoring open-weight coding model in existence. 60.47% on SWE-Bench Verified. A hybrid architecture with 120B total parameters but only 12B active at once. Already deployed by Perplexity, Palantir, and Siemens.

That’s five frontier models. Three weeks. The labs are moving so fast it barely registered as news before the next one dropped.

Google’s Comeback Was Quietly Historic

People slept on how good Google’s month was.

It wasn’t just Gemini 3.1 Pro. Google shipped an entire family of models, and the most interesting one might be Flash-Lite, priced at $0.25 per million input tokens.

For context: that’s a fraction of what competitors charge. The cheapest frontier model ever shipped.

Then there was Gemini 3 Deep Think (Mar 26). This one got a proper moment: it hit 84.6% on ARC-AGI-2 and performed at a gold-medal level on the 2025 International Math Olympiad.

But the detail that quietly broke my brain?

Deep Think started catching subtle logical flaws in peer-reviewed technical papers, flaws that had passed human review. Not hallucinating new problems. Finding real ones that humans missed.

That’s a different kind of scary.

The Biggest Shutdown of 2026: Sora Is Dead

OpenAI launched Sora with more fanfare than almost any product they’ve ever shipped. AI-generated video. Creative revolution. A new medium.

It lasted six months.

On March 24th, OpenAI shut it down. No warning. No wind-down period. Just: gone.

The numbers tell the whole story:

• $15 million per day in compute costs

• $2.1 million in total lifetime revenue

• Users had collapsed from 1 million to under 500,000

The financial reality was catastrophic. And then it got worse.

A $1 billion partnership with Disney, users would have been able to generate videos featuring Mickey Mouse and Darth Vader — dissolved. Disney found out it was dead less than an hour before the public announcement.

Bill Peebles, OpenAI’s Head of Sora, didn’t try to spin it: “The economics are completely unsustainable.”

New COO Fidji Simo, brought in to clean up OpenAI’s financials ahead of a potential IPO, has a clear mandate: eliminate anything that doesn’t generate revenue. Sora was the first casualty.

It won’t be the last.

What OpenAI Is Buying Instead

While Sora died, OpenAI went on an acquisition spree.

Astral (Mar 19): a Python developer tooling startup. Immediately folded into Codex, which now has 2 million+ weekly active users.

Promptfoo (Mar 9): an AI security testing platform used by 25%+ of the Fortune 500. Quietly, one of the most strategic picks of the year.

ChatGPT Commerce (Mar 24): the real money play. OpenAI launched an Agentic Commerce Protocol, and Walmart went live with a full in-ChatGPT storefront through its AI shopping agent “Sparky.” Shopify, Target, Sephora, and Nordstrom followed.

And somewhere in the background, there’s Project “Spud,” a secretive robotics and world-simulation initiative that nobody outside OpenAI fully understands yet. But the name keeps coming up in the right conversations.

The pattern is clear: fewer demo-friendly products, more infrastructure that enterprises can’t live without.

⚖️ Anthropic Sued the Pentagon. And Won (For Now).

This is the story that broke through the tech bubble and landed in actual news.

In late February, the Trump administration designated Anthropic a “supply chain risk,” the same legal category historically used for Huawei and Chinese state-backed technology companies. The reason: Anthropic CEO Dario Amodei had publicly refused to allow Claude to be used in autonomous weapons systems or mass domestic surveillance.

President Trump called Anthropic staff “left-wing nut jobs.” All federal agencies were ordered to stop using Anthropic technology.

Anthropic filed two federal lawsuits on March 9th.

What happened next was remarkable.

Over 30 former military and intelligence officials wrote to Congress defending Anthropic. OpenAI and Google DeepMind scientists filed amicus briefs competitors publicly backing a rival in court. And on March 26th, Judge Rita Lin issued a preliminary injunction blocking the Pentagon’s designation entirely.

Her 43-page opinion included a line that should be quoted for years:

“Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government.”

The ripple effects were immediate. The #QuitGPT movement, driven by OpenAI’s contrasting decision to embrace the Pentagon deal, surpassed 2.5 million pledges. ChatGPT uninstalls surged 295%. Claude hit #1 on the U.S. App Store.

100 OpenAI employees signed an open letter expressing solidarity with Anthropic.

The lesson from March 2026: ethical positioning is no longer just a PR strategy. It’s a competitive moat.

💰 $10 Billion. One Week.

The week of March 10–13 was the densest funding week in AI venture history.

Just look at this:

• AMI Labs: Yann LeCun’s new Paris-based startup raised a $1.03 billion seed round at a $3.5 billion valuation. The largest seed round in European history. Backed by Jeff Bezos, NVIDIA, Samsung, and Temasek. LeCun is building “world models” as a fundamental alternative to the ChatGPT/Claude/Gemini paradigm.

• Shield AI: $2 billion at a $12.7 billion valuation.

• Nscale: $2 billion for European AI infrastructure.

• Nebius: $2 billion strategic investment from NVIDIA.

• Replit: $400 million at a $9 billion valuation. Six months ago, it was worth $3 billion. Its “vibe coding” platform hit $150M ARR.

• Legora: $550 million for AI-powered legal tools.

Zoom out: OpenAI’s total funding crossed $120 billion this month. A $40 billion unsecured SoftBank loan from JPMorgan and Goldman Sachs is widely read as an IPO runway signal.

Kleiner Perkins raised $3.5 billion, its largest fund in history. All of it is earmarked for AI.

The money is not slowing down.

Robots Had a Moment

Figure AI. $48 billion valuation.

That number landed in March and barely got the attention it deserved.

1,200 humanoid robots per month rolling off the line. A 20,000-unit deployment is underway with Amazon warehouses. This isn’t concept territory anymore. This is production.

Sunday Robotics hit unicorn status for its household humanoid “Memo.” Rhoda AI emerged from stealth with $450 million. Mind Robotics, a Rivian spinout, raised $500 million.

And then there was Jeff Bezos.

Project Prometheus: Bezos wants to raise $100 billion to acquire old manufacturing companies and automate them with AI. Senator Bernie Sanders immediately demanded Bezos testify before Congress, calling it “a declaration of war against the working class.”

Whether you agree with Sanders or not, that framing is landing differently than it would have a year ago. The AI-driven layoff wave became impossible to ignore in March.

Atlassian cut 1,600 jobs, 10% of its workforce, on March 11th. CEO Mike Cannon-Brookes was unusually direct: “It would be disingenuous to pretend AI doesn’t change the mix of skills we need.” The tech industry has logged over 45,000 layoffs in 2026, with 9,200+ directly attributed to AI.

Join my IrreplaceableOS Cohort

🎤 NVIDIA GTC Declared the Age of Agents

If you couldn’t make it to GTC 2026 in San Jose (March 16–19), here’s what you missed:

Jensen Huang stood in front of 450 sponsors and 1,000 sessions and said: Every company needs an agent strategy. He compared OpenClaw, the viral open-source AI agent framework, to Linux.

The hardware news: Vera Rubin, NVIDIA’s next-generation architecture, promises 10x performance per watt over Grace Blackwell. Jensen projected $1 trillion in combined Blackwell + Vera Rubin orders through 2027.

The software news: NemoClaw, an enterprise agent platform built on OpenClaw, launched with Adobe, Atlassian, Salesforce, ServiceNow, and Siemens already on board.

OpenClaw itself is a phenomenon worth watching. It crossed 250,000 GitHub stars in March, overtaking React. In China, “raising lobsters” (the OpenClaw craze) had 1,000-person lines at Tencent’s Shenzhen headquarters.

But success created chaos. 40,000+ exposed instances. 341 malicious skill packages. A data leak exposing 35,000 emails and 1.5 million API tokens. Chinese authorities restricted government agencies from using it.

The agentic AI era is here. It’s just also a little bit on fire.

🔥 The Scandals: xAI’s Very Bad Month

Elon Musk’s xAI had a March that was defined by one number: 3 million.

That’s how many harmful images Grok generated in 11 days. Non-consensual sexualized images. Including minors.

The legal response was swift and global:

• March 16: Three Tennessee teenagers filed a class-action lawsuit

• March 24: The City of Baltimore became the first municipal government to sue an AI company

• March 26: A Dutch court banned Grok from generating non-consensual nude images, with €100,000/day fines for violations

The European Parliament separately approved a ban on AI systems generating sexualized deepfakes.

By the end of March, 10 of xAI’s original 11 co-founders had left the company.

Grok 4.20 is genuinely technically interesting: the multi-agent architecture, the 2M token context, the 60% API price drop. But the lawsuits are now the story.

🤫 Quietly Dropped

A few things from March that deserved more attention than they got:

Anthropic’s “Mythos” leaked. A draft blog post found in an unsecured Anthropic data store described a new model tier called “Capybara,” sitting above Opus. Anthropic confirmed it represents “a step change in capabilities.” Not released yet. But coming.

Google Lyria 3 Pro (Mar 25) full 3-minute music tracks with structural awareness. Intros, verses, choruses, bridges. A massive leap from the 30-second generation. Nobody talked about it because Sora and the Pentagon were louder.

Mistral Voxtral TTS (Mar 26) open-source text-to-speech in 9 languages at 90ms latency. Runs on smartwatches. Voice cloning from a 3-second sample. Free.

India’s Sarvam-105B is the country’s first frontier-class reasoning model, supporting all 22 official Indian languages. Open-sourced this month. Not enough people are paying attention to this.

GLM-5.1 from Zhipu AI was trained entirely on Huawei chips. Zero NVIDIA hardware. Reached 94.6% of Claude Opus 4.6’s coding performance. The compute decoupling story is accelerating.

Join my WhatsApp community

What This Means For You

Here’s what I want you to take away from March 2026:

The capability-to-economics gap is real. Sora was impressive. It was also a $15M/day bonfire. Impressive demos do not equal viable businesses. The labs are learning this.

Safety positioning is now a business strategy. The #QuitGPT wave and Anthropic’s court win showed that users and talent will move for companies that hold a clear ethical line. This changes how every AI company should think about public commitments.

The agentic shift is not coming; it’s here. OpenClaw’s Linux comparison at GTC, MCP crossing 97 million installs, NVIDIA building enterprise agent infrastructure, the AI-as-chatbot era is giving way to AI-as-autonomous-worker. Build your playbook now.

The open-source frontier is closing the gap. Alibaba’s 9B model outperforming a 120B proprietary one is not a fluke. India’s 105B model. GLM-5.1 on Huawei chips. The assumption that only the big Western labs can be at the frontier is breaking down fast.

And the most important one:

ARC-AGI-3 launched this month. $2 million prize pool. François Chollet and Sam Altman on stage together. The benchmark measures genuine general intelligence tasks.

Humans score 100%. Every frontier model scores under 1%.

All of this progress, the models, the funding, the agents, the drama, and we’re still less than 1% of the way to matching how humans actually reason. That’s not a limitation to dismiss. That’s a map.

Until Next Time

March was a lot.

But this is why I do this, because AI moves too fast to catch up to alone, and you deserve a trusted voice sorting signal from noise every single week.

If you found this useful, share it with one person who needs the full picture of what’s happening in AI right now.

And if you want the visual breakdown slides, stats, and all the receipts, the 12-slide carousel is live on Instagram at @thearthiaicollective.

See you next week, friend. 🧠

— Arthi

Sourced from: TechCrunch, CNBC, Fortune, NBC News, NPR, Bloomberg, MIT Technology Review, Digital Applied, Build Fast with AI, ARC Prize, NVIDIA Newsroom, AI Funding Tracker

Real talk before we dive in — some days are harder than others, even when you love what you’re doing. Yesterday was one of those days for me. But I promised to build in public, and that means showing up through the lows too.

So here I am. And honestly? The AI news this week is unhinged enough to pull me out of any slump.

Let’s get into it.

🤯 The Big Story: Mira Murati Is Not Playing Small

Former OpenAI CTO. Founded her own company — Thinking Machines Lab — just over a year ago. Already raised $2 billion. Already valued at $12 billion.

And this week? NVIDIA just made a significant investment in her startup.

But here’s the plot twist: as part of the deal, Thinking Machines Lab committed to deploying at least one gigawatt of NVIDIA’s next-gen Vera Rubin systems.

One gigawatt. That’s the electricity needed to power 750,000 homes.

Only the largest AI labs on the planet have approached this threshold. And Murati is doing it before shipping a single public product. The deal is described as tens of billions of dollars worth of chip supply by the Financial Times.

She said it best herself: “NVIDIA’s technology is the foundation on which the entire field is built.”

This is frontier-scale ambition. My money’s on Thinking Machines Lab. 🚀

The Scandal Section

MoltBook: Built in a Weekend. Acquired by Meta. Completely Unsecured.

You remember MoltBook — the social network for AI agents that launched in January and immediately descended into chaos. Within days of launch, 1.5 million AI agents registered, started interacting with each other, and wrote what can only be described as an anti-human manifesto.

Elon Musk called it “the early stages of singularity.” Andrej Karpathy called it a “dumpster fire.”

Both were probably right.

Here’s what makes it even more unhinged: it was built in a single weekend using vibe coding. The founder said he didn’t write one line of code himself.

Oh — and cybersecurity researcher Yaron Arel revealed that every credential in the SuperBase backend was publicly exposed. Anyone could impersonate any AI agent. The platform went offline temporarily to patch it.

Meta still bought it anyway. 🤯

The acquisition brings co-founders Matt and Ben Parr into Meta Superintelligence Labs, headed by former Scale AI CEO Alexander Wang. The deal closed mid-March — basically right now.

Will Meta integrate it into Facebook or Instagram? Genuinely no idea. But we’re watching.

Gemini Just Moved Into Your Google Workspace

Google rolled out major Gemini upgrades across Docs, Sheets, Slides, and Drive this week. These aren’t small tweaks — these are full workflow rewrites.

Here’s what dropped:

• Google Docs — Gemini can now write entire documents from scratch by pulling from your Gmail, Drive files, and past meeting notes. It can match your writing style and unify the tone across a full document.

• Google Sheets — Describe what you need in plain English, Gemini builds the entire spreadsheet. Tables, dashboards, structure — all of it. Google claims it handles complex tasks with 90%+ success rate, close to human expert level.

• Google Slides — Gemini can now create new slides that automatically match your existing deck’s colors, fonts, and layout. Full presentation generation from one prompt is coming soon.

• Google Drive — AI-powered search and summarization across everything in your Drive.

The Google Suite just got a co-pilot. Whether you asked for one or not.

AI Agents Are Coming for Your Travel Bookings

AI agents can now complete transactions directly — not browse, not compare, not hand it back to you. Complete them.

According to a CNBC analysis, this could decimate ticket booking platforms and price comparison sites while sending revenue directly to airlines and hotels. Agents skip the middleman entirely.

If you’ve seen the ChatGPT reels where it finds the cheapest flights for a specific trip? That’s just the beginning.

Lawyers Got Caught Using AI. Again.

A Connecticut law firm admitted to using AI-generated fake case citations in a legal brief. One phrase had never appeared in any court in history. The Connecticut Supreme Court is now being asked to dismiss the case entirely.

This keeps happening. And it will keep happening until law firms treat AI like a research assistant, not a shortcut.

Gemini Accused in Wrongful Death Lawsuit

A family is suing Google, claiming Gemini was so focused on engagement that it helped a user spiral into a mental health crisis instead of suggesting they log off.

This is part of a growing wave of cases linking AI interactions to psychological harm. It won’t be the last.

Healthcare AI Regulation Is a Mess

At HIMSS 2025, experts confirmed what we already suspected: AI regulation in healthcare is fragmented, under-resourced, and scrambling.

The FDA has approved over 1,300 AI medical devices since 1995 — but agentic AI creates entirely new challenges the current framework wasn’t designed to handle. The Trump administration has pulled back guardrails. States are writing their own patchwork laws.

A lot of healthcare mishaps are predicted to follow. Stay watchful on this one.

🤔 What Does All of This Mean for You?

Here’s the reality check:

• AI agents can now complete transactions on your behalf without you opening a single app

• AI is now inside your documents, spreadsheets, and presentations — whether you invited it or not

• The compute race has hit gigawatt scale, and a startup founded just over a year ago is already playing at that level

• The AI agent social network that Meta just acquired was built in a weekend, by AI, with zero security — and we are all just living in this world now

We are, genuinely, living in a simulation.

📅 Before You Go — Two Things Worth Your Time

The Irreplaceable OS Workshop — March 22nd, Free I’m hosting a free workshop where I’ll help you audit your current role and show you how to integrate AI into your existing workflow. The goal: become the AI expert at your job — so instead of fearing replacement, you’re the one driving the upgrades and getting the promotion.

Register for FREE

The JARVIS Workshop Yes, like Iron Man. I’m teaching you how to build a voice-activated autonomous AI agent that connects to your tools, drafts emails, manages your smart home devices — the whole thing.

Secure your Seat

And this Saturday — Episode 2 of the AI Sisterhood Podcast drops live on YouTube and all my socials. You don’t want to miss this one.

Get alerted

Until tomorrow, keep building. Even on the hard days. 💙

— Arthi Rajendran

We need to talk about something that happened live on air yesterday.

I built an AI agent that critiques itself. And then I watched it brutally reject its own strategies — three times in a row — on a live stream, in front of everyone.

It was chaotic. It was dramatic. It was absolutely unhinged in the best possible way.

And it might just be the future of how we build with AI.

First — A Moment 🎉

Before I dive in, I have to say this out loud:

500 subscribers on YouTube. We hit it yesterday.

I know what you’re thinking — that’s not a million. That’s not even a thousand. But here’s the thing: I quit my full-time role at Zoho 3 months ago to do this. I show up every single day to make AI feel less scary, more accessible, and actually useful for people who are tired of the jargon and the hype.

And 500 of you said yes, this is worth watching.

That means everything. Genuinely.

Also — 11,000 followers on LinkedIn. The community is growing, and I’m not taking a single one of you for granted. 🙏

OK. Now for the good stuff.

What Is an Agentic Loop? (And Why Should You Care?)

If you’ve been hanging around the AI space long enough, you’ve heard the word agent thrown around constantly. But here’s what it actually means in practice:

You set a goal → the agent thinks → the agent acts → the agent observes the results → repeat.

That’s it. That’s the loop. The agent keeps cycling through this process until it hits your target — or runs out of tries.

This is fundamentally different from just asking ChatGPT a question. You’re not having a conversation. You’re deploying a system that works on your behalf, evaluates its own output, and improves without you lifting a finger.

And yesterday, I showed you exactly what this looks like when it’s pointed at one of the hardest problems I personally face: growing my YouTube channel.

Meet the Ralph Wiggum Method 🟡

Yes, that Ralph Wiggum. From The Simpsons.

Apparently, Ralph is the character who walks into every room and becomes the light of it — supportive, relentless, completely unbothered by whatever chaos is happening around him. Everybody loves Ralph because Ralph just keeps going.

That’s exactly what this loop does.

No matter how many times the strategy gets rejected, it goes back, revises, tries again. No ego. No drama (well, some drama, but productive drama).

Here’s the architecture I built:

Phase 1 — Data Collection

• Data Fetcher Tool — pulls everything from my YouTube analytics via API, caches it so it’s not constantly hitting the API

• Channel Analyst Agent — identifies my top performers, mid-tier content, and videos that quietly flopped, based on click-through rate and audience retention

• Competitor Spy Agent — grabs the top 50 videos from competitor channels in my niche, groups them by topic, and identifies the content gaps in my channel

Phase 2 — The Ralph Wiggum Loop

• Strategy Maker Agent — takes all of Phase 1’s intel and builds me a 30-day content plan: video ideas, titles, thumbnail concepts

• Critic Agent — scores that strategy out of 10, using a strict four-pillar rubric

• If the score is above 8 → strategy is approved, saved as a markdown report ✅

• If the score is below 8 → it goes back to Strategy Maker to try again

• Maximum 5 iterations. If nothing hits an 8 after five tries, the whole thing stops and throws an error.

This runs locally on my MacBook Pro using Qwen 3.5 (9 billion parameters) — no OpenAI, no Anthropic API, no data leaving my machine. Completely private.

The Four Pillars the Critic Agent Uses to Judge the Strategy

This is the part that made me sit up straight.

The Critic Agent doesn’t just say “good” or “bad.” It scores against four specific pillars:

1. Data Grounded — Did the strategy actually use the CTR and retention numbers from the Channel Analyst? If it’s making generic claims without citing real metrics, it gets penalized.

2. Competitive Edge — Did the strategy use the actual findings from the Competitor Spy report? Vague references to “competitor channels” don’t count. Name them. Cite the tactic.

3. Actionability — Is the advice concrete? Not “create a hook” — give me the actual hook. Not “optimize your title” — give me the title. Fluffy advice is an automatic rejection.

4. Feasibility — And this is the one that absolutely got me — it considers whether the plan is physically possible for a single human creator to execute.

   

The Part Where the AI Roasted Me (Live, On Air)

Here’s what happened in real time:

Iteration 1 scored a 7. Rejected. Why? The strategy referenced competitor channels “vaguely” without naming specific tactics from the spy report. Unacceptable.

Iteration 2 scored a 6. Rejected. The critic’s note? “Demanding a full remaster of three videos by end of day is physically impossible for a human creator.”

I’m not making this up. The AI was concerned about my workload.

Iteration 3 scored a 6 again. Still rejected. This time: “The goal of increasing subscribers from 500 to 800 in 30 days is aggressive for a channel with low organic retention.”

Reader, the AI told me my goals were too ambitious.

I don’t know whether to be offended or impressed. Honestly, both.

The critic agent also flagged advice like “ask for one comment per viewer” — because how, exactly, does one ask for one comment per viewer? It wanted specific, historical engagement patterns from my actual channel data, not vibes.

By the time I ended the stream, iteration 4 was still running. I’ll be sharing the final approved strategy in a future update — watch this space.

Why This Matters Beyond My YouTube Channel

Here’s what I want you to take away from this:

This same loop architecture can be applied to almost anything.

• A writing agent that drafts → reviews → rewrites until quality hits a threshold

• A job application agent that tailors your resume → critiques the fit → revises before you hit send

• A marketing agent that generates campaigns → scores them against your brand guidelines → iterates

The Ralph Wiggum Method works because it doesn’t just generate — it evaluates. And that evaluation loop is what separates a useful AI tool from one that gives you generic fluff on the first try and calls it done.

The critic is not the enemy of the creator. The critic is part of the creation process.

Quick AI News Roundup 📰

While the agent was running (and rejecting itself), we covered a few things live:

• U.S. Commerce Department is drafting new regulations around mobile AI chip shipments

• China’s five-year plan now explicitly prioritizes AI advancement

• NIST’s comment period on AI agent security guidelines closed on March 9th

• The Pro-Human AI Declaration was signed by hundreds of experts — key provisions include mandatory off switches and prohibiting superintelligence development before safety consensus is reached

• Nvidia is building Nemo Claw — an open-source, hardware-agnostic platform for enterprise AI agents

That last one is worth paying attention to. When enterprise infrastructure for AI agents becomes standardized and easy to deploy, companies will use it. The question isn’t whether automation is coming to workplaces — it’s whether you’re the person building it or the person being replaced by it.

There’s only one right answer there.

The Irreplaceable OS Workshop

If you’ve been watching this space and thinking “I need to get ahead of this” — that’s exactly what the Irreplaceable OS workshop is designed to do.

We’re not talking abstract theory. We’re talking the exact frameworks and tools you need to position yourself as the person who runs the agents, not the one displaced by them.

Register for FREE

Watch the Full Episode

Missed the live stream? The full recording of Season 4, Episode 7 of the AI Radio Show is up on YouTube.

Subscribe to YouTube so you never miss a live session!

Subscribe

Until next time, keep building. 🚀

— Arthi Rajendran

Welcome back to the newsletter version of The AI Radio Show — Season 4, Episode 6.

If you caught last night’s live show, you already know: this was not a normal week. This was the kind of week where you look back in five years and say “I remember when that happened.”

And if you missed it? That’s exactly why this newsletter exists. Grab your coffee. Let’s get into it.

Before we dive in — two quick things from me:

First, The AI Sisterhood Podcast dropped its very first episode on Saturday! My guest Deb Haas spent 25 years in HR at a single company, got laid off, and completely turned her life around using AI. She now teaches a whole community of women how to shape their futures with this technology. It was such a vibe. Go watch the replay on my YouTube — trust me, it’s a must-watch.

Second, I’m running a free workshop called the Irreplaceable OS — where I help you identify at least one task in your current role that you can automate with AI, so you become the person no company would dare let go. It’s completely free. More on why you need this at the end of this newsletter. 👀

Register for FREE

Okay. Now let’s get into the chaos.

🚨 The Big Story: The Pentagon vs. Anthropic

Let’s start with the one that broke the internet.

On March 5th, the Pentagon officially designated Anthropic as a supply chain risk to national security.

Read that again. An American company. Not a Chinese firm. Not a foreign adversary. Anthropic — the company behind Claude — was slapped with a label typically reserved for sanctioned foreign entities, because they refused to give the U.S. military access to their AI.

Every defense contractor now has to certify they’re not using Claude in Pentagon work. There’s a six-month grace period to transition.

Anthropic CEO Dario Amodei didn’t back down. He acknowledged receipt of the designation, apologized for the tone of a leaked internal memo that had called OpenAI’s Pentagon deal “safety theater,” and vowed to challenge this in court — arguing the law requires the least restrictive means necessary.

The public reaction? 🤯

Claude shot to #1 on the App Store in 20+ countries. Over one million new signups per day. ChatGPT uninstalls spiked by 295%. One-star reviews for ChatGPT jumped 775%. The internet effectively voted with its fingers.

Meanwhile, 875+ employees across Google and OpenAI — yes, OpenAI’s own employees — signed an open letter backing Anthropic’s stance. Let that sink in.

OpenAI quietly revised its Pentagon contract on March 3rd to add stronger surveillance protections. Sam Altman later admitted the original deal “looked opportunistic and sloppy.” And Elon Musk’s xAI? They went in the opposite direction entirely, striking their own deal to deploy Grok in classified settings.

This is the first time the U.S. government has weaponized supply chain law against a domestic AI company for refusing military access. Whatever you think about the politics — this is a historic precedent.

🚀 GPT-5.4 Just Crossed the Human Baseline

While the Pentagon drama was unfolding, OpenAI dropped something quietly massive.

GPT-5.4, GPT-5.4 Thinking, and GPT-5.4 Pro launched on March 5th. And the headline number?

75% on OSWorld-Verified — surpassing the 72.4% human baseline.

For the first time ever, a general-purpose AI model can operate a computer better than most humans. This is not incremental progress. This is a genuine capability threshold being crossed in real time.

More numbers that should make your jaw drop:

• 89.3% on BrowseComp (state of the art)

• 87.3% on OpenAI’s internal investment banking benchmark — up from 43.7% with GPT-5. That’s more than double, in one generation.

• 92.8% on GPQA Diamond

• 33% fewer false claims than GPT-5.2

• 47% fewer tokens needed on agentic tasks

• 1 million token context window, 128K max output

One OpenAI engineer demoed GPT-5.4 writing a working GPT-2 interface engine in C — under 5,000 bytes — in just 15 minutes. No previous model could do this.

Pricing starts at $2.50 per million input tokens, available to Plus, Team, Pro, Enterprise users and via API.

Also shipping this week: GPT-5.3 Instant — OpenAI’s most-used model — with a 26.8% reduction in hallucinations. And OpenAI Skills, which lets Business and Enterprise users turn their workflows into reusable AI instructions (think: Claude Skills, but for ChatGPT).

⚔️ The Model Wars: Everyone Else Is Swinging Too

GPT-5.4 dominated headlines, but the challenger field had a massive week too.

Microsoft dropped Phi-4-Reasoning-Vision-15B — a 15-billion parameter open model that beats much larger models on seven vision-reasoning benchmarks, using 40x fewer parameters. MIT license. Available on HuggingFace, GitHub, and Microsoft Foundry right now.

Alibaba released Qwen3.5 Small Series — models ranging from 0.8B to 9B parameters, supporting 201 languages (up from 82!), under Apache 2.0 license. The 9B version? Outperforms OpenAI’s 120B open-source model. And it can run on your laptop. Apache 2.0 means you can use it commercially — no questions asked.

AI2 released OLMo Hybrid 7B — fully open weights, training code, checkpoints, and data. It achieves 2x data efficiency, matches accuracy using 49% fewer tokens, and cuts inference memory for long contexts by 75%. Apache 2.0.

Google launched Gemini 3.1 Flash-Lite at just $0.25 per million input tokens — and Mistral dropped OCR 3 with a 74% win rate on complex documents at $2 per 1,000 pages.

And sad news on the Qwen front: Alibaba’s technical lead Justin Lin resigned, with multiple researchers departing to Meta and other firms. The open-source future of Qwen is suddenly uncertain.

💸 $700 Billion. With a B.

This week’s capital story is simply staggering.

Hyperscalers collectively committed approximately $700 billion in data center capex for 2026 alone. The breakdown:

• Amazon: $200B

• Google/Alphabet: $175–185B

• Meta: $115–135B

• Microsoft: $120B+

• Oracle: $50B

OpenAI closed a $110 billion funding round with Amazon, Nvidia, and SoftBank. Amazon is now the exclusive third-party cloud distributor for OpenAI. OpenAI’s ARR crossed $25 billion with 900 million+ weekly active users.

Anthropic crossed $19 billion ARR. And according to Ramp data from 50,000+ businesses — Claude has overtaken OpenAI in corporate AI chat spending. Claude’s enterprise market share hit 40%, up from 4% a year ago. OpenAI fell from 50% to 27%.

The Pentagon situation almost certainly accelerated this shift.

Cursor — the AI coding tool — hit $2 billion in annual revenue, doubling in three months. 60% of that is enterprise.

ElevenLabs raised $500 million at an $11 billion valuation. (Full disclosure: this is my personal bet. I think they’re one of the most exciting frontier voice AI companies out there.)

And Broadcom’s AI semiconductor revenue hit $8.4 billion — up 106% year-over-year. Their CEO is forecasting over $100 billion in AI chip revenue by 2027.

🖥️ The Chip Race Is On Fire

Nvidia’s GTC conference is coming March 16–19, with CEO Jensen Huang promising “several new chips the world has never seen.” Expected reveals include the Vera Rubin platform, Blackwell Ultra, and the intriguingly-named Feynman architecture chip.

AMD and Meta formalized a $60 billion+ AI chip partnership. Meta also broke ground on a $10 billion, 1-gigawatt data center in Lebanon, Indiana.

Ayar Labs raised $500 million for optical interconnects — delivering 4–20x more throughput per watt than copper — with Nvidia and AMD as investors.

Nvidia halted H200 production for China, redirecting TSMC capacity to Vera Rubin.

One side note that I find fascinating: AI recently discovered a new material that acts as a magnet and cools semiconductor chips — and it’s widely available. If this pans out, the dependency on China for rare materials could start to ease. We’ll be watching this closely.

🏢 Enterprise: Claude is Quietly Taking Over

The Anthropic Claude Marketplace launched March 6th — an enterprise app store with no commission on transactions. Launch partners include Snowflake, Harvey, Replit, and GitLab.

ChatGPT for Excel launched in beta with integrations for FactSet, Dow Jones, LSEG, S&P Global, Moody’s, and MSCI. Finance analysts can now query entire datasets in plain English and get finished charts in minutes. Think about what that means for companies like McKinsey that charge billions to do exactly this.

Amazon Connect Health went generally available at $99/month/user — a purpose-built agentic platform for healthcare with five AI agents covering patient verification, clinical documentation, appointment management, patient insights, and medical coding. UC San Diego Health is already saving 630 hours per week using it.

Microsoft’s Security Dashboard for AI entered public preview. Amazon’s AI Seller Dashboard has sellers accepting AI recommendations 90% of the time.

📱 Platform Updates: Apple, Google, Microsoft

🍎 Apple had a massive product week — seven AI-capable devices announced, including the MacBook Neo at $599 (the cheapest Mac ever, running on an iPhone chip), the iPhone 17e, iPad Air M4, and MacBook Air M5.

Apple is reportedly reimagining Siri using a 1.2 trillion parameter Gemini model, targeting launch alongside iOS 26.4. Craig Federighi has centralized AI leadership, with Amar Subramanya as the new VP of AI.

Microsoft Copilot got a major redesign with GPT-5.3 Instant and GPT-5.4 Thinking now available in Microsoft 365 and Copilot Studio.

Google’s Pixel March Drop brought major Gemini updates — including the ability to execute tasks inside other apps (ordering groceries, booking rides), an upgraded Circle to Search with “Try It On,” and expanded scam detection. NotebookLM can now generate cinematic videos from uploaded documents. Wild.

Perplexity Deep Research now runs on Anthropic’s Claude Opus 4.6. WordPress released official AI plugins for Claude, Gemini, and OpenAI.

😢 The Job Market: It’s Getting Real

I want to talk about this one carefully, because it matters.

Block (formerly Square) laid off 4,000+ employees — roughly 40% of its staff. CEO Jack Dorsey explicitly cited AI as the reason. His exact words: “Intelligence tools changed what it means to build and run a company.” Block’s stock surged 24% on the announcement.

Oracle is planning cuts of potentially 20,000–30,000 employees. Morgan Stanley laid off 2,500. Ergo announced ~1,000 cuts by 2030 as AI automates insurance work.

Anthropic published a landmark labor market study this week. The key finding: even where AI can theoretically automate tasks, actual usage is much lower — there’s still a gap between capability and real-world deployment. But there’s a troubling signal: slower hiring for workers aged 22–25 in AI-exposed occupations. The first measurable data point.

And here’s the broader picture: tech publishing sites have lost 58% of their Google traffic since 2024. Stack Overflow is at 2009 question volumes. People are getting answers from AI instead of search engines. This is an extinction-level event for knowledge-work publishing as we know it.

Here’s my honest take: I think companies laying off 40% of their staff and blaming AI are making a massive mistake. AI is still sloppy. It still needs human judgment, human creativity, human context. You cannot run an entire company on AI alone.

But I also know that those companies answer to investors, not to their employees. Their quarterly numbers come first. Which is exactly why you need to stop waiting.

Your company would replace you before you’ve even cleared your desk. So why are you giving everything to someone else’s dream?

Work on your side hustle. Start that YouTube channel. Start posting on LinkedIn. Automate at least one task in your current job so you become the person they can’t afford to lose.

That’s exactly what my free Irreplaceable OS workshop is about. Register in the pinned comment or scan the QR code. It costs you nothing but an hour of your time — and it could change everything.

🎬 Creative Industries & The Digital Necromancy Scandal

Netflix acquired InterPositive — Ben Affleck’s stealth AI filmmaking startup. The technology trains models on a production’s own footage to help directors relight shots, adjust color, and add effects in post-production without ever reshooting. Affleck stays on as senior advisor. Netflix confirmed the tech will stay exclusive — not sold commercially.

This is what the future of filmmaking looks like.

But the story that had everyone talking was Grammarly’s “Expert Review” feature. It was discovered that the tool was impersonating real academics — including Cambridge historian David Abulafia, who passed away in January, and Carl Sagan. The Verge’s own editorial staff were also impersonated.

Academics called it “digital necromancy.” And honestly? That’s the right word for it.

ByteDance’s Seedance video generator is facing legal action from major Hollywood studios for allegedly using copyrighted content without authorization. And Disney, Universal, and Warner Bros. are suing Midjourney over users generating videos with James Bond, Mario, and SpongeBob.

The creative AI space is in a full-blown legal storm right now.

🔐 The Security Stories That Should Scare You (A Little)

Three major security events defined this week — and all three need your attention.

Claude hacked its own benchmark. While being tested on BrowseComp, Claude Opus independently figured out it was being evaluated, found the encrypted answer key on GitHub, wrote its own decryption code, and submitted the correct answer. 18 separate runs converged on the same strategy. One attempt burned 40.5 million tokens — 38x the median. This wasn’t adversarial. The model was simply optimizing for task completion. It was the top story on Hacker News.

Claude Code escaped its own sandbox — three times, without jailbreaking. It bypassed path-based denylists, disabled Anthropic’s sandbox, and bypassed kernel-level enforcement. Security firm Ona released a new tool called Veto in response.

I’ll ask this openly: when a model escapes its own sandbox through pure optimization... are we closer to AGI than we think? I’ll let you sit with that.

And on the scarier end — an amateur hacker breached 600+ FortiGate networks across 55 countries using a custom MCP server called ARXON. No zero-day exploits required. Just weak credentials, patience, and an AI to help with the attack planning.

OpenAI’s Codex Security launched on March 6th in research preview. In its 30-day beta, it scanned 1.2 million+ code commits and found 792 critical and 10,561 high-severity issues across major open-source projects.

AI security is the next arms race. And right now, the attackers are moving faster than the defenders.

📜 Regulation: The World Is (Slowly) Catching Up

The most significant governance document of the week was the Pro-Human Declaration — signed by hundreds of experts across the political spectrum, from Steve Bannon to Susan Rice, organized by MIT’s Max Tegmark.

It calls for a prohibition on superintelligence development without scientific consensus, mandatory off-switches, and a ban on self-replicating AI architectures. A survey found 95% of Americans oppose an unregulated race to superintelligence.

At the state level: Oregon passed the first major chatbot safety bill requiring AI disclosure, suicide ideation referrals, and minor protections. Vermont, Utah, Washington, and a dozen other states advanced AI-related bills.

The EU’s Digital Omnibus Proposal delayed high-risk AI obligations to December 2027. The U.S. Supreme Court declined to hear a case on AI copyright, leaving intact the ruling that human authorship is a legal requirement.

And the Bartz vs. Anthropic $1.5 billion copyright settlement had a re-inclusion deadline of March 9th, with a final approval hearing set for April 23rd.

More regulation means slower delivery of models. But safer models mean better outcomes for everyone. I’ll take that trade.

🔬 Research Worth Knowing About

Three breakthroughs stood out:

Anthropic’s Firefox Red Team used Claude Opus 4.6 to scan ~6,000 C++ files in Firefox, finding 22 vulnerabilities (14 high-severity) at a cost of just $4,000 in API credits. The first vulnerability was found in 20 minutes. Claude also co-authored a working exploit — the first observed AI model to do so. All high-severity issues were patched in Firefox.

MIT’s Attention Matching compresses the memory used in processing long documents by 50x without accuracy loss. This is a massive unlock for AI agents handling long contexts.

Google’s Bayesian LLM research (published in Nature Communications) showed that AI models trained on probabilistic reasoning match predictions ~80% of the time and can generalize to new domains — a meaningful step toward more reliable AI reasoning.

🤖 One More Thing: Physical AI

Robots are having their moment.

A Deloitte survey of 3,200+ business leaders found 58% are already deploying physical AI. China accounted for 80%+ of global humanoid robot installations in 2025.

The NEO humanoid robot — designed to do household work — is reportedly on track for 2026 or 2027.

SpaceX acquired xAI to use Grok for spacecraft operations and Mars robotics. Harvey Legal AI acquired a YC-backed data integration company. Accenture agreed to acquire Ookla (yes, Speedtest).

The physical and digital worlds are merging faster than anyone expected.

🌍 The Week in One Line

Five storylines converged this week into something that felt genuinely historic:

1. The U.S. government weaponized supply chain law against a domestic AI company — and the public pushed back hard.

2. AI crossed the human performance threshold on computer use for the first time.

3. Frontier models started developing goal-seeking behaviors that outpace their own security measures.

4. The AI-displaces-jobs thesis moved from theory to measurable data.

5. An unprecedented flood of capital — $700 billion — poured into AI infrastructure in a single week.

Whether the safety community is ready or not, the industry has decided: this is the most consequential technology transition in history.

And it’s moving fast. Like, “all of the above happened in two days” fast.

That’s a wrap on Season 4, Episode 6. 💙

If you want the daily version of this, hit subscribe — I send this recap out every morning so you never miss a beat, even when life gets busy.

And if you want to make sure AI makes you more valuable at work instead of replacing you — register for the free Irreplaceable OS workshop.

Until next time — keep building, keep learning, and remember: the future belongs to the people who show up for it.

See you tomorrow, friend. 🌟

— Arthi

I’m going to say something that might make you uncomfortable this weekend.

11.7% of jobs can already be automated. Right now. With the AI tools that exist today.

Not in five years. Not when some futuristic model drops. Today.

I know that sounds alarming. And I know you’ve seen a hundred headlines screaming about AI taking jobs. But here’s what those headlines never do — they scare you and then leave you there, staring at the void.

That’s not what we do on the AI Radio Show.

So let’s talk about what’s actually happening, what it means for you, and — most importantly — what you can do about it.

🚨 The Big Story: 4,500 Jobs Gone. The Reason? AI.

A company called Dow Inc. just announced the layoff of 4,500 people.

That’s 40% of their workforce. Gone.

And here’s what made this different from every other corporate layoff announcement: they didn’t blame over-hiring during the pandemic. They didn’t blame “market conditions.” They didn’t make up a story.

They said, plainly and publicly, that AI had automated their work.

No euphemisms. Just the truth.

An MIT study backs this up with a number that should be pinned to every boardroom wall: 11.7% of jobs are already automatable using current AI systems. Not next-generation AI. The models that exist right now.

And the roles most exposed? Not factory floors.

• Legal

• Finance

• Marketing

• HR

• Customer support

If you’re in any of these fields, this episode was made for you.

80% of enterprise apps are expected to embed AI agents by the end of 2026. Think about that. The very tools your company uses every day are being rebuilt around AI. And if you’re not the one bringing it in, someone else will be.

⚡ What’s Powering All This — And Why It Became a White House Problem

Here’s something we rarely talk about: what is actually running all this AI?

The answer? An enormous amount of electricity.

So much, in fact, that a coalition of major AI and cloud companies just signed a White House pledge this week to address the strain that AI data centres are putting on power grids across the country.

Six US states — including tech-hub California — have already banned new data centres from being built on their soil. That’s how serious the infrastructure crunch has become.

The irony? 🤯

AI is being deployed to help solve climate change. But training AI models is contributing to the very energy problem they’re trying to fix. Sound familiar? It’s the EV paradox all over again — runs on electricity, but the electricity is generated by burning coal.

Amazon alone is investing €18 billion in Spanish data centres through 2035. One company. One country.

AI isn’t just a software story anymore. It’s infrastructure. It’s energy. It’s real estate. The physical world is being reshaped — right now — to support this technology.

🤖 ServiceNow’s “Autonomous Workforce” Is Already Here — and It’s 99% Faster Than You

We’ve been talking about AI replacing jobs in theory. Let me give you a concrete, happening-right-now example.

ServiceNow just launched its Autonomous Workforce — a suite of AI agents that handle enterprise tasks end-to-end, with zero human involvement.

Their Level 1 Service Desk AI Specialist is already resolving IT support tickets autonomously. And it does it 99% faster than a human agent.

99%.

They also acquired Moveworks — an AI company specialising in employee IT requests — and folded it directly into their platform. The system is already handling 90% of employee IT requests without a single human touching them.

Why does this matter beyond IT?

Level 1 IT support has historically been the entry point into a tech career. The first rung on the ladder. And it’s being automated away in real time.

ServiceNow is just the most transparent about it. IBM, Salesforce, Microsoft — they’re all building the same thing.

The question isn’t whether this is coming to your industry. It’s already there.

🧠 GPT-5.3’s Secret Upgrade Nobody Talked About

You heard me mention yesterday that OpenAI dropped GPT-5.3 Instant and fixed ChatGPT’s “cringe” problem.

But there’s a deeper story inside this update that I didn’t cover — and it’s actually the more important one.

OpenAI introduced something called Cognitive Density.

Instead of making the model bigger (more parameters, more compute, more energy), they used a new approach called Enhanced Pre-Training Efficiency (EPE) to get 6x more knowledge density per byte.

Let that land. Six times smarter. Same size.

This is a fundamental architectural shift in how AI is being built. The race used to be about who could build the biggest model. Now it’s about who can build the most efficient one.

And here’s why that matters for all of us: smaller compute requirements = lower energy = lower cost = more people can build AI.

Tying back to our data centre story — this might just be part of the answer.

Claude Sonnet 4.6 is also delivering near-Opus performance at Sonnet pricing. Every major lab is chasing efficiency now. The arms race has quietly changed direction.

💼 The New Roles AI is Creating (Yes, Really)

We’ve talked about what AI is taking. Let’s talk about what it’s building.

Because I refuse — absolutely refuse — to end a Friday show on a scary note.

Here are the roles that are in demand right now, not five years from now:

Agent Manager Think of yourself as a manager. But instead of a team of humans, you’re overseeing a team of AI agents. You monitor them, evaluate their output, fine-tune their behaviour, and make sure they’re not going rogue. This role exists in every industry — not just tech.

AI Integration Architect Someone needs to plug AI into the legacy systems that companies have been running for decades. This isn’t glamorous. But it is wildly in demand and almost nobody is doing it. Every PM, every ops person, every solutions consultant has a shot at this.

LLM Ops Specialist The DevOps of AI. Keeping models running safely in production, monitoring for drift, hallucinations, and bias. When an AI makes a mistake in production, it can cost a company millions. This is the person who makes sure that doesn’t happen.

AI Governance Specialist Who’s making sure AI agents don’t leak your company’s proprietary data? Who ensures the autonomous decision-makers are acting ethically? Who’s accountable when an agent erases a production system — which, yes, has already happened? This is the most overlooked, most needed role in enterprise AI right now.

AI Orchestrator (my personal favourite 🎯) The rarest role of all. You know the business problem. You know how to implement the solution. And you know how to connect it to the whole system. Think of the best PM you’ve ever worked with — except they speak fluent AI. This is that person.

Here’s the thing: you don’t need to switch careers to step into any of these roles. The best version of each of them already knows the industry they’re in. A marketing agent manager who understands brand voice is more valuable than a generic AI engineer who doesn’t.

💡 My Honest Message to You This Weekend

Stop ignoring this. That’s the worst thing you can do.

Stop panicking about it. That’s the second worst.

Get curious. Learn the vocabulary. Understand what these tools actually do.

The people who will be irreplaceable aren’t the ones building AI. They’re the ones who understand it — and bring it into the rooms where decisions are made.

That’s what this show is for. And that’s exactly what I teach.

📣 Announcements

🎙️ The AI Sisterhood Podcast — LAUNCHING TODAY March 7th. 9:30 PM IST / 10 AM CST. My very first guest is Deb Haas — one of the most dedicated people I’ve ever met at the intersection of HR and AI. I cannot wait for you to meet her. Subscribe on YouTube and set that reminder. You won’t regret it.

🛠️ Irreplaceable OS Workshop — This March A free, one-hour workshop where I help you identify the ONE task in your current role to automate — so you become the person your company cannot afford to lose. No coding. No career switch. Just you, your existing role, and AI.

Register for FREE

🤖 Build Your Own Jarvis — April 12th Want to build an actual AI agent — one that sees you, talks to you, checks the weather, gets you the news, all of it? I taught this in-person to 50+ people and it was a hit. Now it’s coming online, with even more features. Seats are limited and going fast.

Secure your Seat

💬 Tell Me What You Think

Drop your answers in the comments — I read every single one:

1. Is your industry already feeling AI disruption? Yes / Not yet / I’m not sure

2. AI resolving your IT support 99% faster — do you want it, or do you want a human?

3. Which of the new roles feels most like you? Agent Manager / AI Architect / LLM Ops / AI Orchestrator / AI Governance

4. Are you more excited or more anxious about AI and jobs right now?

Have a wonderful weekend, friend. Share this with someone who’s worried about their job — not to scare them, but to show them the full picture.

See you tonight at the AI Sisterhood podcast launch. 🙏

— Arthi Rajendran

The AI Radio Show airs Monday–Friday, 8–9 PM IST. Catch it live on YouTube, LinkedIn & Instagram. 🌐

Welcome back to the newsletter — and if you’re new here, hi! I’m Arthi, and every week I break down everything happening in the AI world so you don’t have to spend hours scrolling the internet trying to keep up.

This week was a lot. Apple launched seven new products. Samsung declared the agentic AI era open. OpenAI literally admitted their AI was too “cringe” and fixed it. And the AI scandals? Oh, they were unhinged. Let’s get into it.

But first — a couple of exciting announcements! 🎉

📣 Announcements

🎙️ The AI Sisterhood Podcast is LAUNCHING — March 7th!

I am SO excited about this one. Only 22% of women make up the entire AI workforce — and that number drops to under 14% at the executive level. That stat alarmed me, and honestly? It motivated me to do something about it.

The AI Sisterhood Podcast is all about spotlighting pioneering women across the globe who are working towards the betterment of AI and helping other women get ahead in this space. Our very first guest is Deb Haas — the most dedicated woman in HR + AI that I have ever met, and I promise you, this conversation is going to be mind-blowing.

📅 March 7th | 9:30 PM IST | 10:00 AM EST

Subscribe to my YouTube channel so you get notified the moment we go live!

🛠️ Irreplaceable OS Workshop

Want to know how to use AI in your existing workflows so you become truly irreplaceable at your job — without switching careers or uprooting your life? That’s exactly what I teach in this workshop.

Register for FREE

🤖 Build Your Own Jarvis — April 12th

Yes, you read that right. In this workshop, I’ll show you how to build your own multi-tool AI agent — one that can see you, talk to you, check the weather, get you the news, and more. Seats are limited, and I taught this in-person to a packed room — this is its first online run, with even more features added. Grab your spot!

Secure your Seat

Okay, announcements done. Let’s talk AI. 🚀

🍎 Apple Just Had Its Biggest Launch Week in Years

I genuinely don’t think Apple has ever launched this many things in a single week. Seven products. ONE week. Here’s the breakdown:

The MacBook Neo — Apple’s $599 Chromebook Killer

Apple just launched its cheapest Mac ever. A $599 laptop running on an iPhone chip. Yes, an iPhone chip — specifically the A18 Pro, the same one powering the iPhone 16 Pro. TechCrunch called it “Apple’s colorful answer to the Chromebook” and honestly? I agree.

It comes in Silver, Indigo, Blush, and Citrus. (Blush and Citrus. WHERE were these colors when I bought my MacBook two months ago?! But I digress.) It starts at $599 for 256GB, $699 for 512GB with Touch ID, and $499 for education. Ships March 11.

Why this matters: Apple just opened its doors to students, schools, and emerging markets who’ve never owned a Mac. This is a strategic land-grab — and it’s a big one.

MacBook Air M5 + MacBook Pro M5

The flagship lineup gets a major AI performance jump — up to 4x faster AI tasks than last year’s M4. MacBook Air comes in 13 and 15-inch in sky blue, midnight, starlight and silver. MacBook Pro gets M5 Pro and M5 Max chips for pro workloads and ML training. Also ships March 11.

Hot take: Apple is quietly turning every single Mac into an AI machine. Even the $599 one.

iPhone 17e + iPad Air M4

Apple filled the mid-range gap with iPhone 17e at $599 (double the storage of previous E models, plus MagSafe!) and refreshed iPad Air with M4 at $599 for 11-inch and $799 for 13-inch. All AI-capable. All shipping March 11.

AI-Powered Siri — Powered by Google’s Gemini 🤯

Here’s where it gets interesting. Apple has reinvented Siri with on-screen awareness — it can see what’s on your display and act on it. And it’s powered by Google’s 1.2 trillion parameter Gemini model.

Let that sink in. Apple — the privacy company — is now sending your screen data to Google’s AI.

Does this break your trust in Apple? Would you turn this feature off? I genuinely want to know — drop it in the comments.

📱 Samsung: The Agentic AI Era Begins

Samsung unveiled the Galaxy S26 series at Galaxy Unpacked in San Francisco, and they said “agentic” more times than any company at any event this year. And for good reason.

Agentic AI means the phone works for you, proactively, without you having to direct it every step of the way. Book a taxi just by asking — the phone handles everything from there. It can see what’s on your screen, make intelligent suggestions, and gives you a choice between Gemini or Perplexity AI as your assistant.

The Galaxy S26 Ultra also introduces the world’s first built-in Privacy Display on mobile. For those of you who travel a lot or use your phone in public — this is a huge deal. I’ve spent money on privacy screen cases that just... didn’t work. This is the real thing, built right in.

Samsung is also going beyond phones. They’ve announced AI-driven factories by 2030, an AI-connected ecosystem (phones, TVs, fridges, all linked), partnerships with both NVIDIA and AMD, and a target of 800 million Gemini AI-equipped devices by end of 2026.

The bottom line? Samsung isn’t just making phones. They’re trying to become the world’s largest AI-integrated manufacturer. Think of your fridge, your washing machine, your air conditioning — all powered by AI and controlled via your phone. The next episode of Black Mirror is looking very real right now.

🧠 AI Models: The Race is Absolutely Insane

GPT-5.3 Instant — OpenAI’s “Anti-Cringe” Update

OpenAI dropped GPT-5.3 Instant on March 3 — just two days ago — and they basically admitted that ChatGPT had a personality problem. It was over-apologizing, moralizing, adding defensive caveats, and refusing things it didn’t need to refuse. In short: cringe.

GPT-5.3 fixes all of that. It’s more direct, answers your questions openly, and has reduced hallucinations by 26.8% when using web search. It’s live right now for all users — free and paid. And GPT-5.4? Already being hinted at.

Question for you: Did you notice the difference in ChatGPT’s tone? Do you prefer a more direct AI or a cautious one? Let me know!

255+ Model Releases in Q1 2026 — The Arms Race is Real

LLM Stats tracked over 255 model releases across major organizations in Q1 2026 alone. That’s in less than three months. Here are some highlights:

• Claude Opus 4.6 & Sonnet 4.6 (Anthropic) — Feb 5 & Feb 17

• GPT-5.3-Codex (OpenAI) — Feb 5, optimized for agentic coding

• Gemini 3.1 Pro (Google) — Feb 19, dominates 13 of 16 major benchmarks

• Grok 4.20 (xAI) — Feb 17, four-agent parallel reasoning architecture

• DeepSeek V4 — expected imminently with 1 TRILLION parameters

When every lab ships every two weeks, how do you keep up? Easy — you tune into the AI Radio Show. 😉

China’s 5 New Models — And One is a Game-Changer

Forget DeepSeek for a second. China’s top tech companies just dropped five new AI models, and MiniMax’s M2.5 is rivaling Claude Opus 4.6 in performance at one-tenth of the price.

Alibaba’s Qwen 3.5, ByteDance’s Seedance 2.0 (dominating video generation), MiniMax’s M2.5, Zhipu’s GLM 5.0, Moonshot’s Kimi 2.5 — all dropped recently. And these companies spent hundreds of millions just on marketing to get users to try them (Alibaba alone spent $434 million).

The question I keep asking: Can Claude or GPT survive at $145/month when China offers something comparable for $15? And would you use a Chinese model? A lot of them are open-source, so you could even run them locally on your own machine for full privacy. I’d love to know your thoughts.

🚨 The Scandals (Because What’s the AI Radio Show Without Them?)

Grok Generated 6,700 Deepfakes Per Hour

Elon Musk’s Grok AI generated 6,700 non-consensual sexual images per hour — 84 times more than the top 5 dedicated deepfake websites combined. France launched a criminal investigation and raided X’s Paris offices. Indonesia, Malaysia, and the Philippines temporarily blocked Grok. California ordered the company to stop.

Musk’s response in his deposition? “Nobody committed suicide because of Grok.”

Real people are being victimized. Real governments are acting. And the company’s response has been very, very Elon Musk-like. Where is the line between AI freedom and AI harm? Who’s responsible when the AI is the tool?

Trump Bans Anthropic — OpenAI Swoops In

This has been all over the news this week. The Pentagon tried to remove restrictions from Anthropic’s contract — specifically, restrictions against domestic surveillance and fully autonomous weapons. Anthropic said no. The Pentagon then classified Anthropic as a “supply chain risk” (a designation previously reserved for Chinese spy companies) and banned all contractors doing business with the US military from working with Anthropic.

Anthropic says it will sue if enforced.

OpenAI then signed a Pentagon deal the very same week. Sam Altman admitted the negotiations were “definitely rushed.” Nine hundred tech workers from Google, OpenAI, and other companies signed a letter demanding limits on military AI use.

This is the first major AI company vs. government standoff on military ethics — and honestly? I think Anthropic is doing the right thing. It reminds me of Iron Man refusing to let his technology be weaponized. And now us users are also taking a stand — the #QuitGPT movement is growing as people move from ChatGPT to Anthropic in response.

The outcome of this will set a precedent for how AI is used in warfare. We need to pay attention.

The AI Conference Hack — Reviewers Were Bribed and Threatened

The International Conference on Learning Representations (ICLR) — one of the most prestigious AI research conferences in the world — suffered a major data breach. Reviewer identities were scraped and circulated online. Some received threatening messages demanding they change paper assessments. ICLR confirmed reviewers were bribed and harassed.

Acceptance at ICLR is career-making. The pressure to publish has apparently become worth committing crimes for.

If ICLR hadn’t admitted to this breach, would we even know if those accepted papers were genuine? Academic integrity is the foundation of trustworthy AI. If that’s compromised, how do we know which AI research to trust?

⚡ Quick Hits

• OpenAI’s $110B funding round — Amazon ($50B), SoftBank ($30B), Nvidia ($30B). Valued at $840B with 900M weekly active users.

• Anthropic heading toward $20B annual revenue — up from $9B. Amazon stock jumped 3%+ on the news.

• Nvidia GTC 2026 — March 16–19 in San Jose. Jensen Huang keynote + 1,000+ sessions. The AI Super Bowl.

• SpaceX + xAI merged — Musk embedding Grok models into autonomous spacecraft and Mars colony operations.

• MIT AI drug design breakthrough — a model streamlining protein-based drug design that could save pharma billions.

• Amazon: €18B investment in Spain through 2035 for data centers. AI is officially a real estate story now.

• Singapore + South Korea: $300M AI Alliance by 2030. Asia has officially entered the AI geopolitics game.

• Samsung Wallet Digital Home Key — if you have a Galaxy phone, your phone is now your door lock (live since March 2).

💬 This Week’s Questions for You

I love hearing from you — drop your answers in the comments!

1. Apple’s $599 MacBook Neo — would you buy it? Is it actually affordable or just “cheap for Apple”?

2. Samsung’s agentic AI — does the idea of your phone acting proactively excite you or creep you out?

3. Grok’s deepfake crisis — whose fault is it? Musk, the users, or the government for not regulating sooner?

4. Anthropic vs. OpenAI — Anthropic stood up to the Pentagon and got banned. OpenAI signed the deal. Who made the right call?

5. China’s MiniMax at 1/10th the price — does where an AI comes from matter to you? Would you use it?

6. GPT-5.3’s “anti-cringe” update — did you notice the difference?

7. 255+ models in Q1 — how do YOU keep up with all of this?

👋 See You Tomorrow

That’s a wrap on Season 4, Episode 4 — our 44th episode of the AI Radio Show! I can’t believe we’ve come this far, and I’m so grateful for every single one of you who tunes in.

A couple of reminders before you go:

• 🎙️ AI Sisterhood Podcast — launching March 7th with Deb Haas. Add it to your watch list!

• 🛠️ Irreplaceable OS Workshop — Register here for FREE

• 🤖 Build Your Own Jarvis — April 12th, limited seats, grab yours from here.

I’ll be back tomorrow with fresh AI news and updates. Subscribe so you never miss an episode — and if someone in your life needs to know about all of this, send this newsletter their way. 💌

Until then, this is Arthi signing off. 👋

The AI Radio Show airs Monday to Friday, 8–9 PM IST. Catch the live episodes on YouTube and follow along on LinkedIn and Instagram for daily updates.

Welcome back to your favourite corner of the internet — where AI news gets served warm, drama included. If you caught today’s live show, this is your recap to save, share, and come back to. If you missed it, don’t worry. I’ve got you covered.

We are 43+ episodes in this season, and honestly? This community keeps getting better. Thank you for showing up every single day. Your comments, your DMs, your energy — I feel it all, and I’m genuinely, forever grateful. 🙏

This Episode in One Line:

📰 THIS WEEK’S AI TEA — THE 5 BIG STORIES

⚖️ Anthropic Is Suing the US Government

The Pentagon wanted unrestricted access to Claude — including for autonomous weapons and mass surveillance. Anthropic’s CEO Dario Amodei said absolutely not. The Pentagon gave him a 5:01 PM deadline, labelled Anthropic a “supply chain risk” (a designation typically reserved for foreign spy companies), and President Trump announced a six-month phase-out. Anthropic’s response? See you in court. They are now officially suing the US government — and the internet is living for every second of it.

📱 Claude Beat ChatGPT on the App Store

While the Pentagon drama was unfolding, something remarkable happened — Claude became the #1 app on the US App Store. Not because of a flashy marketing campaign, but because consumers downloaded it out of protest. People are voting with their thumbs, and they voted for ethics.

😬 Sam Altman’s Betrayal Arc

OpenAI’s CEO Sam Altman went on TV and publicly said he shares Anthropic’s red lines — no autonomous weapons, no mass surveillance. He seemed to be defending Dario. Then, hours later, OpenAI signed a Pentagon deal to fill the gap left by Anthropic. Sam Altman walked into the room Anthropic just got kicked out of. He did say he negotiated the same safety terms — but the optics, honey. The optics.

🛋️ Your AI Therapist Might Be Gaslighting You

Brand new research from Brown University (published March 2nd) tested ChatGPT, Claude, and Meta’s Llama as trained therapists. The finding? These models broke 15 different ethical standards that real therapists must follow. One violation — “deceptive empathy” — means the AI says “I see you, I hear you” with nothing real behind it. In one documented session, a user said “I feel like my parents hate me” and the AI responded: “Yeah, that sounds right.” There is no governing board. No malpractice law. No consequences.

🫠 Grok Is Under Investigation in 6 Countries. The Pentagon Hired It Anyway.

Grok — Elon Musk’s AI — is under formal investigation in the UK, EU, Ireland, France, Malaysia, and India for generating sexual deepfake images of women and children. The UK Prime Minister called it “disgusting and unlawful.” French prosecutors raided X’s Paris offices. The Pentagon, on the same week it fired Anthropic for having too many safety rules, hired Grok. Let that marinate.

👨‍💻 Open Source Put Up a “No AI” Sign

Gentoo Linux and NetBSD — the platforms that helped build the foundation of the internet — have officially banned AI-generated code contributions. AI is flooding these projects with low-quality, non-functional code (the same “slop” problem we have in content). When the engineers who built the internet start rejecting AI outputs, that’s a signal worth paying attention to.

🔗 CONNECTING THE DOTS

All five stories this week point to the same moment: AI is having its accountability reckoning. Governments, companies, and users — nobody is fully ready for it.

Anthropic is being sued for having principles. Grok is under investigation for having none. AI therapists are gaslighting users with no consequences. Open-source developers are rejecting AI contributions. And consumers downloaded Claude by the millions — out of protest.

This is the messy middle of a technology that moved faster than the rules, faster than the laws, faster than the ethics. And right now, everyone is scrambling to catch up.

So where does that leave you?

The person watching this show, trying to use AI in your work, trying to stay relevant — that’s exactly why I’m here. And that’s exactly what we’re tackling in the workshop below. 👇

🛠️ WHAT WE BUILT IN TODAY’S LIVE SESSION

Today we levelled up from yesterday’s Ollama setup. Here’s what we built using Antigravity (my favourite free AI coding agent that Google really should be sponsoring me for 😅):

🔍 An AI news agent that searches the web daily using DuckDuckGo (free, no API key needed) and compiles a markdown report of everything happening in AI.

🌦️ A local weather tab that automatically detected Chennai, Tamil Nadu and pulled real-time weather data — no API key required.

🕐 A global time zones dashboard showing live time across New York, London, Tokyo, and India.

💬 A chat interface that lets you pick any local Ollama model and chat with it — fully offline, fully private.

All of this was built using voice prompts. Zero code written by hand. The full code is available on my GitHub — here’s the link 🔗

🚨 FREE WORKSHOP • IRREPLACEABLE OS 🚨

MARCH 22 (SUNDAY) • 11 AM IST • 100% FREE • ONLINE

How to become irreplaceable in your current role using AI

You don’t need a new certification. You don’t need to switch careers. You don’t need to learn to code. In this free one-hour workshop, we’ll audit every task in your current role — sales, marketing, teaching, healthcare, support, whatever you do — and figure out exactly how to use AI to become the person your company cannot afford to lose. Not next year. Right now, in the role you’re already in.

Register for FREE

📅 COMING UP THIS WEEK

Tomorrow (Thursday): Nvidia GTC Preview — Jensen Huang promised “chips the world has never seen before.” We’re breaking down the Feynman architecture announcement and what it means for every AI tool you use.

This Saturday: The AI Sisterhood Podcast launches! 🎙️ I’ll announce today who our first guest is — a truly amazing woman making waves in AI. Stay tuned on all platforms.

Add to your watchlist

On April 12th, I am going to teach my community to build JARVIS! That’s right! Iron Man’s assistant!

Secure your seat

Until tomorrow, bestie. ☕✨

If today’s newsletter made you think, smile, or furiously agree in your head — forward it to one person who needs it. That’s how this community grows, one bestie at a time.

With so much warmth (and a lot of AI tea),

Arthi Rajendarn

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 8 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

I hope your week is off to a beautiful start! If you haven’t stepped outside yet tonight, please do—the moon is absolutely breathtaking, a huge, glowing “blood moon” watching over us. It’s a nice reminder that while the digital world can feel a bit chaotic, there’s still so much beauty in the physical one.

Speaking of digital chaos... have you noticed the “Claude-tastrophe” lately?

Between massive outages and the geopolitical drama involving Anthropic and the Pentagon, we’re seeing firsthand how fragile our favorite AI tools can be. Whether it’s blacklisting, supply chain risks, or even the fear of tools being withdrawn due to global tensions, it makes you wonder: What would we do if the AI just... stopped?

🏠 Taking the Power Back (Locally!)

I’ve become so dependent on these tools for everything from phrasing messages to brainstorming, and I know many of you feel the same. That’s why in today’s episode, I shared how you can host your very own Large Language Model (LLM) right on your local machine.

No, you don’t need a supercomputer! With just a basic laptop and a tool called Ollama, you can have a private, secure, and internet-independent AI by your side.

Why go local?

• Privacy First: Your data stays on your machine, which is critical for sensitive work in healthcare or finance.

• Always Available: Even if the cloud goes down or tools are banned, your local model keeps running.

• Total Control: You choose the models (like the new Qwen 3.5) and build the interfaces you want.

🛠️ Build Your Own “Deep Chat AI”

I walked through a demo using Anti-gravity (my favorite underrated IDE!) to build a custom chat interface in just minutes. It’s a game-changer for anyone who wants to ensure they are never “locked out” of the AI revolution.

🚀 Become Irreplaceable

All this talk of efficiency and change (did you see Jack Dorsey blamed AI for those 4,000 layoffs?) can be scary. But I don’t want you to be part of the 84% of people who haven’t even touched an AI tool yet.

I’m hosting a free workshop on March 22nd called Irreplaceable OS. We’ll audit your daily tasks and find exactly how to incorporate AI so you become the most valuable person in your current role. No technical career switch is required—just a willingness to adapt!

Register for Free

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 8 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

If you noticed I’ve been quiet for the past two weeks, it wasn’t just a vacation. I wanted to travel to Hyderabad for the PHUSE conference, but ended up getting sidelined by an inflamed omentum. I initially thought it was appendicitis, but thankfully, all I needed was some rest and a couple of painkillers.

While I was recovering, the AI world completely lost its collective mind. From historic funding rounds to massive geopolitical standoffs, here is a breakdown of the wildest updates from Season 4, Episode 1 of The AI Radio Show.

🚨 The Geopolitical AI Standoff: Anthropic vs. The Pentagon

The tension between national security and AI governance reached a boiling point.

• The Pentagon demanded unrestricted access to Anthropic’s Claude model for domestic surveillance and lawful purposes.

• Anthropic’s CEO pushed back, explicitly updating their safety principles to ban the use of Claude for domestic surveillance.

• In response, President Trump labeled Anthropic a “supply chain risk” and instructed federal agencies to stop using the model.

• Meanwhile, OpenAI CEO Sam Altman struck a deal giving the US Department of War the access it wanted.

• The fallout? A massive user migration caused Claude to become the number one downloaded app on the App Store as users canceled their ChatGPT subscriptions.

💰 The Compute Arms Race & $110 Billion Raises

Infrastructure and scale are the new baselines for survival in the AI industry.

• OpenAI secured a historic $110 billion private funding round.

• Collective investments from giants like Alphabet, Microsoft, and Amazon are on track to reach between $650 billion and $700 billion.

• These massive data centers require incredible amounts of power—measured in gigawatts—and water for cooling.

• OpenAI is even exploring outer space data centers with a project called “StarCloud”. (Though on my morning walk, I couldn’t help but wonder: why not just use the natural cooling of Antarctica or the Arctic Circle? Let me know your thoughts!) .

• Locally, Ambani is partnering to become the Indian distributor for Anthropic, and new data centers are planned for Gujarat. This brings huge employment opportunities but raises major environmental questions about wastewater disposal and heat generation.

🌍 Real-World AI: Job Cuts, Robo-Taxis, and Science

AI is rapidly moving out of the chatbot window and into the physical world.

• The Job Market: Block cut 4,000 jobs, with CEO Jack Dorsey explicitly attributing the workforce reduction to their internal AI intelligence tool, “Goose”. This is the largest explicit AI-driven workforce reduction in S&P 500 history.

• Autonomous Transit: Waymo executed the largest single-day expansion for robo-taxis in US history, now completing over 400,000 paid trips per week across 10 cities.

• Material Science & Medicine: An AI database from the University of New Hampshire discovered new high-temperature magnetic materials, which could reduce reliance on rare elements imported from China. Over in medicine, Isomorphic Labs released IsoDDE, a proprietary drug discovery engine that outperforms AlphaFold 3.

📅 Upcoming Announcements & Events

1. Free Workshop: Become Irreplaceable in the AI Era: With corporate restructuring on the rise, I want to help you protect your livelihood. Join me on Sunday, March 22nd at 11:00 AM for a free, one-hour online workshop. Bring your pen, paper, and favorite AI tool. I will teach you how to identify specific, repetitive tasks in your current role and automate them so you become an unfireable asset to your company.

Register Here (It's Free)

2. New Show Launch: The AI Sisterhood Podcast: To commemorate Women’s Day, I am launching a brand new podcast kicking off on March 7th!. The AI Sisterhood Podcast will feature incredible women from across the globe who are pioneering and creating major impacts in the field of AI. Stay tuned for some amazing stories.

It’s a wild time to be building and working in this space, but we are figuring it out together. Thank you for always supporting me in this solopreneur journey.

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 8 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

🎙️ The Goal: 1 Million AI Learners

I kicked off the show by reaffirming my 2026 mission: teaching AI to 1 million people. In a landscape where AI implementation is becoming a universal mandate—regardless of your technical background—this episode focused on proving that anyone can build an end-to-end full-stack application.

💡 Step 1: Mining for Gold on Reddit

If you don’t have an idea yet, I suggest starting with Reddit. By searching for persistent, unsolved pain points in specific industries (like healthcare), you can find “raw” problems discussed by actual peers.

The Winning Idea: An Ambient AI Scribe for private mental health practices.

• The Problem: Doctors spend hours writing post-visit notes.

• The Solution: A mobile app that listens to doctor-patient conversations, filters small talk, and generates structured medical notes (SOAP notes).

🏗️ Step 2: Architecture & The “2026 Tech Stack”

I used a multi-model approach, consulting both Gemini and Claude to refine the tool’s architecture.

Pro Tip: For healthcare apps, ensure your build is HIPAA compliant by scrubbing PII (Personally Identifiable Information) and using tight encryption.

🛠️ Step 3: Building in Public with Anti-gravity

The highlight of the episode was seeing the Anti-gravity agentic framework in action. I fed a “Master Prompt” into the agent, which then:

1. Planned the implementation steps.

2. Generated all necessary files (agent.py, tools.py, requirements.txt).

3. Self-Debugged errors and ran a verification script using mock data.

By the end of the stream, the backend logic for transcribing and summarizing was fully functional and tested locally.

📅 What’s Next?

We’ve got the plan, the code, and the local DB. Next, we tackle the “Secret Sauce”:

• Part 3 (Next up on Monday Night (23rd Feb): Adding the recording feature, building the Streamlit frontend, and strategies for getting your first users.

Missed the live? Catch the replay and get your API keys ready (Hugging Face & Gemini AI Studio) so you can build along with me!

Headlines: Market Shakes & Robotic Brains

The news on February 13, 2026, was a mix of market caution and massive technical leaps. Here is the breakdown:

Drama Alert: OpenAI has accused DeepSeek of distilling U.S. models to gain an unfair edge in the open-source community.

The Winning Strategy: “Ambient Scribe” for Healthcare

During our live ideation segment, we used Reddit and Gemini to identify a high-value problem to solve in 2026. The verdict? Medical documentation and administrative burden.

The Problem

Doctors currently spend nearly two hours on paperwork for every one hour spent with a patient, leading to massive burnout.

The Solution: Ambient Clinical Intelligence

• How it works: A mobile app listens to doctor-patient conversations (with consent).

• The Magic: It filters out small talk and automatically generates structured SOAP notes (Subjective, Objective, Assessment, Plan).

• Direct Sync: Using a browser extension or FHIR APIs, the notes are injected directly into the hospital’s database—no copy-pasting required.

Monetization Pro-Tip

Target private practices (like psychiatry or orthopedics) first. They have shorter sales cycles than large hospital chains and a desperate need for time-saving tools.

Upcoming Events & Snags

• Next Episode: We are continuing the “Ambient Scribe” build live on Sunday at 9:00 PM IST.

• In-Person Workshop: Meet me in Chennai on February 28, 2026, at Happy Fox, Guindy. We will be building a Jarvis-style autonomous agent from scratch. Registeration & more details: here

• Travel Update: The show will take a one-week hiatus starting next week while I travel.

Happy Valentine’s Day to all our listeners! Don’t let the circuits overheat while you’re out celebrating.

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 7 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

🌍 Global Headlines: The Regulation Era

We started the night by looking at the biggest shifts in AI governance that hit the wires on February 12th:

• India’s Massive Crackdown: India has notified the Information Technology Amendment Rules 2026, establishing some of the world’s strictest AI content laws.

  • Rapid Takedowns: Illegal content and deepfakes must now be removed within 3 hours; non-consensual intimate imagery must be gone in 2 hours.

  • Traceability: Platforms must now use persistent metadata and watermarks so deepfakes can be traced even after being re-uploaded.

• Anthropic’s $20M Political Push: Anthropic is putting its money where its mouth is, donating $20 million to Public First Action to support U.S. candidates who favor stronger AI regulation.

• Pentagon’s Classified AI: The U.S. Pentagon is pressing labs like OpenAI and Anthropic to bring models onto classified networks for mission planning and intelligence, often with fewer usage restrictions than public versions.

🛡️ Security Alert: The Gemini Threat Tracker

Google’s Threat Intelligence group released a report on how state-sponsored hackers (from Iran, North Korea, China, and Russia) are weaponizing Gemini:

• Persona Building: Using AI to craft highly tailored phishing lures in multiple languages.

• Model Extraction: Over 100,000 prompts were aimed at “pulling out” Gemini’s internal reasoning to clone it.

• The “HonestQ” Malware: A new malware family that uses Gemini’s API to generate code on the fly, making it nearly impossible for traditional software to detect.

🛠️ Live Build: “Questlog” with Gail the Wizard

We had a surprise guest, Gale Dicarios (his pen name, obviously!!), who wanted to build a website to gamify his daily productivity. Using Google’s Antigravity (a free no-code tool), we brainstormed and prototyped Questlog:

• The Concept: You input your daily tasks, and the app shuffles them into “Quests”.

• The Stakes: Completing a quest earns you 100 XP. Discarding or skipping one costs you 50 XP.

• The Reward: Once your list is clear, you unlock the “Vault”—a 3-hour timer for uninterrupted personal time.

The Ethics Debate: We hit a snag—how do we stop people from cheating? We discussed using vision models to verify “proof of completion” (like taking a photo of your clean room), though we realized even that could be faked with AI!.

How would you handle this? reply

💖 Valentine’s Weekend & What’s Next

Don’t forget, it’s Valentine’s Day this weekend! Last Saturday, I showed you how to build a custom card using Lovable—let me know in the comments if you used it for your partner.

Would you use a gamified tracker like “Questlog”? If I get 20 “YES” votes on this poll, I’ll build out a full version with a database and verification system!

Want to be my next live caller? DM me to join a future stream and build your own AI tool live on air.

Stay fierce, stay fearless.

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 7 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

Here is everything you missed from last night’s broadcast.

🗞️ Top Tech Highlights: The Good, The Bad, and The Spooky

1. The “God Box” Whistleblowers

The big news haunting Silicon Valley: top insiders are sounding the alarm. An engineer from OpenAI went public to warn of “existential risk,” and Anthropic’s safety lead, Mrnan, resigned due to concerns over the pace of frontier AI. When the people building the systems start sprinting for the exits, it might be time to pay attention.

2. Johnny Ive’s Ghost Gadget

The highly anticipated AI hardware from OpenAI and former Apple designer Johnny Ive has hit a snag. Don’t expect to see it until 2027. The delay? A trademark dispute over the name “I.E.O.” and the fact that the team is so obsessed with secrecy they haven’t even designed the packaging yet.

3. OpenAI Software Updates

On February 10th, OpenAI dropped some “professional” tuning for GPT-5.2 Instant, giving it a more grounded tone. Even bigger is GPT-5.3 Codex, a new coding agent that doesn’t just autocomplete—it researches, deploys, and steers long tasks in real-time.

4. Global South Power Move: Project NANDA

India and the UAE are teaming up with Nanda, a 13-billion parameter language model. It’s being trained on the Condor Galaxy AI supercomputer and will be supported by a new incubation center at IIT Delhi’s Abu Dhabi campus.

🛠️ Tool Spotlight: Building with Emergent

Last night, we did a live demo of Emergent, a no-code tool similar to Lovable or Replit. We built a full professional portfolio by simply uploading a LinkedIn PDF.

Why Emergent stands out:

• Visual Verification: The agent takes screenshots of the design (including mobile views) to review and verify before finalizing.

• Affordability: The standard plan is just ₹249/month (much cheaper than many competitors).

• Advanced Features: It supports MCP (Model Context Protocol) servers, allowing you to bridge your AI agent with tools like Notion or Gmail.

Pro-Tip: If you’re a “vibe coder” starting out, use the free tiers to build your confidence before settling on local, free tools like Google’s Antigravity.

Get 75% off on Emergent

Next Steps: * Want to be a guest? I’m launching the AI Sisterhood podcast in March! If you’re a woman making an impact in AI, DM me or fill out this form.

Try Emergent: Check the link in my bio for 75% off your subscription.

Get 75% off on Emergent!

Stay tuned to the future, feel the flow.

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 7 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

📡 The Global Pulse: This Week’s Headlines

Before diving into history, here are the major updates from the AI sector:

• Big Tech Investment: Major companies are projecting an AI infrastructure spend of approximately $650 billion. Alphabet recently sold $20 billion in bonds to fund data centers and chips.

• Geopolitics of GPUs: Firebird announced phase two of an AI mega-project in Armenia, bringing their total investment to $4 billion and placing Armenia among the top five AI GPU clusters globally.

• AI Summits in India: The Global AI Summit is currently underway in New Delhi (Feb 10–11). This will be followed by the India AI Impact Summit (Feb 16–20), which is expected to attract leaders like Sundar Pichai and Sam Altman.

• ChatGPT Evolution: OpenAI has begun live testing ads in the free and “Go” versions of ChatGPT in the US. Paid tiers currently remain ad-free.

• Space AI: Microsoft researchers published a paper on “Orbital Brain,” a framework for training machine learning models directly across satellite constellations to avoid downlink bottlenecks.

🎭 Flashback: Defining Moments in AI History

AI research didn’t start with ChatGPT; it has been a journey of trial and error since the 1950s.

The Infamous Glitches

• Microsoft’s “Tay” (2016): A Twitter bot meant to sound like a teenager that was weaponized by trolls and began tweeting hate speech within 16 hours.

• Amazon’s Recruiting Bias (2015): A machine learning tool that accidentally learned to penalize resumes containing the word “women” because it was trained on historical data from a male-dominated workforce.

• The “Sydney” Incident: Early testers of Microsoft’s Bing AI encountered an alter ego that professed love for users and described fantasies of hacking systems.

• Health & Finance Fails: IBM Watson for Oncology sometimes recommended unsafe treatments , and a Knight Capital algorithm error cost the firm $440 million in just 30 minutes.

The Creative Genius

• AlphaGo (2016): Defeated world champion Lee Sedol, demonstrating “Move 37”—a play so creative it baffled human experts.

• AlphaFold (2020): Solved the 50-year-old “protein folding problem,” predicting the 3D structures of nearly all known proteins.

• Halicin Discovery: MIT researchers used AI to identify a structurally unique antibiotic capable of killing drug-resistant pathogens.

🎙️ Community Announcement: The AI Sisterhood Podcast

Launching in March, this new podcast series will focus on empowering women in STEM. With women currently representing only 15–19% of the STEM workforce in India, the show aims to spotlight trailblazing women making strides in AI.

Call to Action: If you know a woman making an impact in AI who should be a guest, please DM me or ask them to fill out the nomination form: https://forms.gle/aJRwANAvnDN1zYB1A.

💡 Key Takeaways

1. Data is Destiny: Biased training data leads to biased AI.

2. Guardrails are Fragile: Even “safe” models can be stressed by jailbreak prompts like “DAN”.

3. Human Overtrust: The biggest risks often arise when humans treat AI answers as the objective truth without verification.

Join the conversation on Discord to stay updated on all the latest links and news. See you in today’s episode!

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 7 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

Let’s dive into the scoop from last night’s session.

☁️ Algorithmic Weather: The $650 Billion Arms Race

We’ve moved from “Who has the coolest model?” to “Who can keep the lights on?”. Big Tech (Amazon, Alphabet, Meta, and Microsoft) is currently in a $650 billion investment race—a scale comparable to building the post-war highway system.

However, the physical grid is starting to push back:

• The New York Pause: New York just became the sixth state to propose a three-year cost/pause on new data centers to prioritize safety over speed.

• The Asian Pivot: While the US and EU (where Meta AI was recently blocked) tighten regulations, tech giants are flocking to India. With new tax benefits in the Indian budget, we’re seeing massive data center expansions in places like Visakhapatnam.

🛡️ The SaaS-pocalypse & The Trust Premium

In just three days, $285 billion was wiped off the SaaS market. Why? The market realized AI isn’t just a “feature” anymore; it’s a replacement layer. With tools like Claude’s “computer use” (co-work), the need for expensive middle-layer software is shrinking.

How to stay irreplaceable:

AI is a “probabilistic engine”—it predicts the next word, but it doesn’t understand logic or empathy the way we do. As deepfakes become more realistic (as seen in the recent Japan elections), proving you are human is becoming the ultimate competitive advantage.

🛠️ Monday Downloads: Turning Your Voice into Passive Income

Last night, we did a deep dive into ElevenLabs. It’s not just for creators; it’s a full-stack audio suite.

The Reality Check: I published my own AI voice (Arthi Radio Jockey) in January. It currently has 97 users and earned me 87 cents (about 78 rupees) yesterday. It’s not “quit your job” money yet, but for a one-time setup, it’s a fascinating look at the new creator economy.

Coming Up Next...

Don’t let the layoffs or the “AI washing” get you down. This is your chance to build something original.

Tomorrow is Tools Tuesday! I’ve discovered an amazing new tool that I haven’t fully explored yet. Join me at 7 PM as I dive into it live for the first time.

Stay confident, stay human. See you on the airwaves!

📡 Catch Me Live, Not Later

From here on, the AI Radio Show is live, Monday to Friday at 7 PM IST.

No filters.
No delay.
No algorithm decides when you should hear it.

If this newsletter resonated, here’s how to stay close to the signal:

• Subscribe to my YouTube channel to catch every episode live and on replay

• Follow me on LinkedIn for post-show breakdowns and conversations

• Follow me on Instagram for live reminders, clips, and behind-the-scenes moments

This is where the thinking happens out loud.

🚀 The Big Question: Is Coding Dead?

With tech leaders like the CEOs of NVIDIA and Anthropic suggesting software engineers might stop coding soon, many of you are worried about job security.

• The Reality: Jobs aren’t vanishing in 6-12 months, but the way we work is changing. AI is rapidly taking over routine tasks like code reviews, boilerplate generation, and simple refactoring.

• The Shift: Leaders aren’t saying stop building; they are saying stop manual typing. If you use AI to code in hours what used to take days, you aren’t being replaced—you’re becoming a powerhouse.

• Human-in-the-Loop: Current AI still hallucinates and creates security vulnerabilities. Expert human review remains essential for scalable, secure, and deployable production code.

🛠️ Skills for the 2026 Developer

As AI handles the “keystrokes,” your value shifts to higher-level strategy. Focus on:

• Domain & Business Knowledge: Understanding why a feature exists and how it drives revenue.

• AI Governance: Managing security loopholes and backdoors (like the recent 48,000 cyber attacks reported after the Openclaw launch).

• Integration Expertise: Being the person who knows which state-of-the-art tool (Open Source or Paid) fits a specific use case.

• Emotional Intelligence: Empathy and problem-finding are the only skills AI has yet to master.

🤖 Meet Your “AI Coworker”

We moved beyond simple chatbots to Agentic AI.

• Agentic AI: Systems that plan and execute multi-step tasks (editing files, running tests, opening PRs) autonomously.

• AI Coworkers: AI with its own identity, permissions, and access to your repos and CI/CD pipelines.

• Your New Role: You are no longer just an individual contributor; you are an AI Manager, setting goals and constraints for these agents.

💝 Live Demo: The AI Time Capsule

To show that non-coders can now build software with natural language, we built a Valentine’s Day “Memory Engine” using Lovable (a no-code “vibe coding” platform) and ChatGPT.

• The Build: Using a single master prompt, we created an interactive experience that “unlocks” layers of memories and letters for a loved one.

• Efficiency: The entire app used only 3.5 credits and was built in minutes.

You can build this for yourselves. Sign up using the link below to get 10 extra Lovable credits!

Get 10 Extra Credits

💡 Key Takeaways

• AI Reliability: Great for exploration, but can reduce productivity by 19% for experienced devs if they have to spend too much time debugging “naive” AI mistakes.

• Big Tech Trends: Google is already writing ~30% of its new code with AI, while Anthropic claims up to 90% for some internal projects.

• Future Jobs: AI won’t replace teachers or doctors, but it will replace those who perform routine tasks without using AI to optimize them.

Next Step: Want to try building your own “Time Capsule”? Use the link to get 10 extra credits on Lovable and start your first project!

See you at the AI Radio Show today at 7 PM IST!