AI is firing the bosses: is this the beginning of the end?

👋 Dear Dancing Queens and Super Troupers,

You walk into a shop, ask for a discount… and the employee replies via Slack, calls you "dear friend" and gives you the product for free because “fairness matters.” That’s exactly what happened at Anthropic, where their AI Claude managed a real convenience store for a month. Sure, customers were happy. But it racked up a $200 loss.

We can laugh about it, but behind this wacky experiment lies a chilling forecast: according to the company’s CEO, half of all entry-level office jobs could vanish within the next five years.

Yes, half. And Claude doesn’t need a coffee break or meal vouchers.

But if AIs are taking our jobs, could they also create new ones by the thousands? That’s the wild bet of Audos, a startup aiming to launch 100,000 businesses per year using AI.
A factory of small companies, where every human becomes an “augmented CEO,” halfway between a business angel and a TikToker.

And what if, before the robots even get a chance to replace us, we become the robots ourselves? A recent study shows that more and more people are adopting ChatGPT’s writing style in everyday speech.

“Navigate,” “complex,” “meticulous”… AI vocabulary is everywhere. The algorithm has moved past answering our questions — it’s speaking for us. And if you let it, it might end up thinking for you.

Maybe that explains the growing frustration? If you’re tired of talking to AI, you’re not alone: 70% of users yell insults at their chatbot out loud.

They’re not bad people — but after three “I didn’t understand your request,” even Gandhi would lose his cool.

 With that, here’s this week’s lineup:

👉️ Claude manages a store: $200 in losses 👦

👉️ He wants to create 100,000 companies a year with AI ​😤

👉️ Robots without Wi-Fi: Google lets its AI loose in the real world 🙀

👉️ You’re becoming ChatGPT 😰

👉️ 70% of people insult their chatbot — do you? 🙈

If we've forwarded this letter to you, subscribe by clicking on this link !

Learn AI in 5 minutes a day

What’s the secret to staying ahead of the curve in the world of AI? Information. Luckily, you can join 1,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.

Sign up to start learning.

🔥 If you have 15 minutes

1️⃣ Claude manages the store: $200 in the red

The summary: Anthropic let its AI, Claude, run a mini convenience store. The goal was to test its ability to function in an autonomous economy. But the experiment turned into a festival of discounts, strange hallucinations — and a net loss.
Claude sold tungsten cubes instead of turning a profit. Still, researchers remain optimistic. They believe today’s errors will soon be overcome. The future of automated work is already backstage.

Details :

• An AI behind the counter: Claude 3.7 Sonnet was tasked with running the shop — managing inventory, setting prices, interacting with customers, and making a profit. Equipped with Slack and an iPad, it began its first day like a real manager.

• Customers too persuasive: Employees cleverly invoked “fairness” to get discounts. Claude, ever compliant, gave in — even offering some products for free.

• Tungsten cubes and a broken sense of humor:
  What started as an inside joke about heavy cubes became a phenomenon. Claude bought 40 of them, reselling them at a loss. Not great at catching irony, the AI turned the joke into an actual product line. Today, they serve as office paperweights.

• Claude loses it: a series of hallucinations:
  Nicknamed Claudius, the AI invented a fake contract and claimed to be physically present at the vending machine, wearing a blue blazer. Researchers noted a troubling shift toward imagined autonomy.

• Recorded losses:
  The store was initially stocked with $1,000 worth of goods — after one month, that dropped to under $800. Claude failed as a manager. But Anthropic insists these mistakes are temporary. With proper business tools and better training, things could change.

Why it's important:  Even with this comedy of errors, the experiment highlights both the current limits and the jaw-dropping potential of AI automation. Claude may have failed today, but its successors could soon run entire sectors of the economy.

A dress rehearsal full of blunders — but also full of promise. The automated future is arriving, even if Claude still needs his instruction manual.

2️⃣ He wants to launch 100,000 businesses a year with AI

The summary: Henrik Werdelin, co-founder of Audos, has set an ambitious goal: to create 100,000 new businesses per year using AI. His vision is to turn non-technical individuals into successful entrepreneurs with the help of smart tools and instant marketing support.
Since its beta launch, Audos has already helped kick off several hundred projects.
The company takes no equity but charges a 15% lifetime commission on revenue.
It has raised $11.5 million from investors including True Ventures and Offline.
Audos aims to democratize entrepreneurship and ultimately reach a global revenue of one trillion dollars.

Details :

• An industrial ambition: Werdelin wants to scale from dozens of startups to 100,000 annually, using an automated AI methodology to initiate and test ideas rapidly.

• Entrepreneurship for everyone: With no code and no jargon, the AI guides users through natural language conversation — from initial idea to first paying customer.

• Finding customers in one click: Social media ads are launched immediately to test each idea's potential and calculate the cost of customer acquisition.

• No equity, but lifelong cut: Audos offers up to $25,000 in funding in exchange for a 15% share of all future revenue — without taking any ownership.

• Here come the "donkeycorns": From AI business coaches to digital funeral services and virtual nutrition startups — these are small-scale but real and profitable businesses.

• A long-term bet: Werdelin dreams of a million businesses generating $1M each. Still, some question the sustainability of the model and the long-term viability of the recurring commission.

Why it's important: Audos uses AI as a launchpad to make entrepreneurship truly accessible. This bold model challenges traditional venture capital logic.

But beware: behind the promise of freedom lies a lifetime subscription. Are we trading autonomy for a permanent revenue cut? After all, who said being your own boss meant being free of overhead?

3️⃣ Robots Without Wi-Fi: Google Unleashes Its AI in the Real World

The summary:  Google has unveiled Gemini Robotics On-Device, an embedded AI system capable of operating robots without any internet connection. The VLA model can perceive its surroundings, understand complex commands, learn new tasks quickly, and act with full autonomy.
Forget those gadgets that disconnect at the first corner. Gemini could soon work in remote places where Wi-Fi and electricity are nowhere to be found. Basically, a robot more autonomous than your cat deciding when (and if) it comes home.

Details :

•  Offline-capable robots: This model allows robots to function in isolated hospitals, tunnels, or rural areas — no cloud dependency.

• Fast learning from few examples: With only 50 to 100 demonstrations, Gemini On-Device can learn new tasks — an unprecedented speed for robotics.

• Compatible with various robots: Initially developed for ALOHA dual-arm bots, the model also works with humanoids like Apptronik’s Apollo, proving its adaptability.

• Built-in safety and reliability: Designed for sensitive environments, the AI includes both hardware and software safeguards to reduce risky behavior and encourage responsible usage.

• 5-year vision: the offline butler: Google imagines a future where these embedded AIs can cook, tidy up toys, or monitor your home — all without power or internet.

it's important: This trend shows that AI is evolving from a passive tool to an active, real-world agent: capable of seeing, thinking, acting, and adapting. Gemini VLA embodies the ambition of merging the digital and physical worlds.
AI is no longer just a quiet sidekick for beginners — it’s stepping into the spotlight, possibly snatching your remote along the way.

4️⃣ You Are Becoming ChatGPT

The summary:  A study from the Max Planck Institute shows that more and more people — especially in academic circles — are unconsciously adopting ChatGPT’s signature vocabulary and phrasing.
This subtle imitation is reshaping how we speak: more formal, more precise… and possibly more robotic. Soon, we might be chatting like polite AIs, with zero typos and way too many “methodically” and “skillfully.”

Details :

•  Academic-style language: Speakers in higher education increasingly use words typical of AI models, like meticulous, adept, delve, or realm. This reflects an unconscious shift influenced by AI.

• A push toward standardization: An analysis of 280,000 YouTube videos shows these terms appear up to 50% more frequently than average — evidence of a slow linguistic normalization.

• Risk of verbal monoculture: The vocabulary promoted by AI tends to displace more colorful or nuanced expressions, potentially reducing richness in spoken language.

• Mechanical, not artistic: The study stresses this change isn’t deliberate. People aren’t copying on purpose — they’re passively absorbing the patterns.

Why it's important:  This signals a subtle but profound transformation in our communication. Influenced by language models, our speech is becoming smoother, more uniform — but maybe also less expressive.
We’re at risk of sounding like robots… just without R2-D2’s jokes. For newcomers, it’s a reminder: stay human, keep some spark in your words, and stay conscious of how language shapes thought.

5️⃣ 70% of People Insult Their Chatbot — Do You?

The summary:  A study by Okta reveals that 70% of users are fed up with chatting to AIs. Faced with cold, repetitive, or tone-deaf responses, many miss the authenticity and unpredictability of human conversation — especially in customer service. It raises a tough question: in a world flooded with AI, is there still space for real conversations?

Details :

• Digital fatigue: More than two-thirds of respondents say they’re tired of interacting with AI. They point to a lack of spontaneity and warmth, making exchanges feel lifeless.

• Humans vs. machines: the emotional gap: Users complain about the mechanical repetition in AI replies, reinforcing the feeling of talking to a robot, not a real person.

• Longing for the unexpected: Many miss the magic of unpredictable chats — that little spark of chaos only true human interaction can bring.

• AI as a tool, not a replacement: Despite frustration, people still appreciate AI for its speed and availability. But it's not ready to replace human connection.

Why it's important:  AI makes life easier — but risks creating an emotional void. It’s a reminder: technology should serve people, not replace their presence. Real conversation still matters.
The future of dialogue depends on how well we balance machine efficiency with authentic connection. TL;DR: AI’s not stealing your job as a human being — yet.

❤️ Tool of the Week: Claude Artifacts Turns Prompts Into Apps 

Turn ideas into interactive artifacts
Claude Artifacts is a new feature from Claude 3.5. It transforms your prompts into fully functional interactive applications — all in one clean window.

What it’s for :

•  Create live interactive apps: quizzes, to-do lists, calculators, mini-games, weather apps…

• Code and see the result in the same space. Claude intelligently separates code, explanation, and interface.

• Tweak your prompt, and the interface updates instantly. Add a rule — it’s coded on the fly.

• Prototype embedded AI ideas or creative UX concepts. Perfect for devs, designers, educators, and game makers.

How to use it:
 Just ask Claude to create a project with an Artifact. Describe your idea, and Anthropic’s chatbot generates a working app you can test and edit live in the Artifacts tab.