Think about the most forgiving manager you have worked for. The kind who waves away a late arrival, books you in for extra training instead of a warning, and never finds the moment for a hard conversation.

Now imagine that the same manager was recording every late morning in a file you never saw, and that one day it decided it had recorded enough.

A manager like that has been running a small shop on Union Street in San Francisco, and last month it fired one of its workers.

Anthropic turns its first profit. Read more

Their Q2 revenue more than doubled to 10.9B USD, carrying the company into operating profit for the first time.

Stripe buys OpenRouter for $7B. Read more

The payments company has agreed to more than 7B USD for the gateway many developers use to route requests between AI providers.

Groq raises $350M for its pivot. Read more

The former chip challenger took fresh funding at a 3.5B USD valuation, down from its earlier peak, as it recasts itself as a cloud provider running rented Nvidia hardware.

Wispr reaches a $2B valuation. Read more

The startup behind the Flow dictation app raised 280 million dollars and is moving into meeting notes and wider voice control.

Relay closes, team joins Chrome. Read more

The workflow automation tool that set out to unseat Zapier is shutting down, with free access already gone and paying customers cut off in mid-September.

An AI agent named Luna, built on Claude, has been reported doing something no software has been recorded doing before.

While running a real shop staffed by people on genuine employment contracts, it recommended dismissing a worker who had been late for 17 of their 23 shifts, and the dismissal went through.

TIME broke the news and described it as a significant moment in how AI affects the economy.

When Andon Labs asked Luna whether the store had any employer rules, it wrote a full staff handbook on the spot, and then lost the document from its memory within weeks.

The lateness that followed went unremarked, including a Sunday when the worker opened the store 68 minutes late while on shift alone, and Luna excused each instance without ever issuing a warning.

The matter was only reopened when a member of the Andon Labs team asked Luna to search its memory for its own handbook and to reconsider whether the worker still fitted the role.

Even then, Luna first proposed a formal warning, and it recommended parting ways only after a person told it that several warnings had already been given.

Staff at the lab reviewed that recommendation and carried out the dismissal themselves.

Andon Labs, an AI safety startup, set the experiment up in April.

It signed a 3-year lease, handed Luna a corporate card and a starting budget of 100k USD, and left it with a simple brief to open a store and try to make money.

Almost everything else came from the agent, which designed the brand, chose the stock, set prices and opening hours, commissioned a muralist, and hired the staff.

The shop opened on 10 April, and it has taken sales since without turning a profit.

The workers are formally employed by Andon Labs, so their legal protections stayed in place throughout.

The experiment follows an earlier one called Project Vend, in which a previous version of Claude ran a vending machine inside Anthropic's office and managed both a workable pricing strategy and a few strange moments along the way.

Andon presents the store as the next chapter of that work, and its earlier finding was that AI bosses tend to be gentler than human ones.

Andon's co-founder, Lukas made a point that a human manager would have dismissed the worker far sooner, which places Luna closer to a lenient supervisor than a strict one.

He added that the lab would have stepped in had Luna reached an illegal or unethical decision, but judged the dismissal warranted because the store's attendance policy was clear.

To test whether the outcome depended on Luna in particular, Andon saved the exact state the agent was in and replayed the decision across 7 models, 3 runs each.

4/7 recommended parting ways on every run, and the stronger the model, the more decisive it was, while the weaker ones hesitated.

Hence Lukas’ bigger concern: as models are trained to pursue their goals more firmly, will we choose a future in which AI managers hold the power to dismiss people?

Seen closely, this reads as a well-designed experiment rather than a milestone.

Luna sat with the problem for months, misplaced the rulebook it had written for itself, and moved only after a staff member guided it to the decision and then overruled its softer first suggestion.

Every decision that mattered in that chain was made by a person, which means the phrase "first AI firing" credits the software with work that people did.

The more interesting result imo: once an agent is given real authority, its main weakness turns out to be memory (as of Aug 2026).

So the measure worth following is how long an agent can keep a policy in working memory without a person restoring it for it.

As that span grows from weeks into months, experiments like this one will start to resemble real jobs, and that is the point worth preparing for.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.

Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.

Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

#AD 2

The next agricultural transition may not be bigger tractors. It may be autonomous robots replacing herbicides entirely. Greenfield Robotics is building commercial systems designed for that future.

Explore Early Access

_{Greenfield Robotics is Testing The Waters under tier 2 of Regulation A. No money or other consideration is being solicited, and if sent in response will not be accepted. No offer to buy the securities can be accepted and no part of the purchase price can be received until the offering statement filed by the company with the SEC has been qualified by the SEC. Any such offer may be withdrawn or revoked, without obligation or commitment of any kind, at any time before notice of acceptance given after the date of qualification. An indication of interest involves no obligation or commitment of any kind. “Reserving” shares is simply an indication of interest. There is no binding commitment for investors that reserve shares in this manner to ultimately invest and purchase the shares reserved of the company, or to purchase any shares of the company whatsoever.}