A creator publicly begging employees of a top AI lab to quit looks like drama, and the open letter to Anthropic exactly that on the surface. Underneath, it raises one testable question: is Anthropic shipping features faster than it can keep them stable?
What Does the Open Letter to Anthropic?
The open letter to Anthropic a public post in which the creator Theo Browne urges engineers at Anthropic, the company behind Claude, to leave the lab. He claims the CEO disrespects engineers, that bad code reaches users, and that staff stay only for unvested equity. He offers to help recruits land better-paid roles elsewhere.
The letter's core offers are concrete. The author says he has friends at most AI labs and startups, that many would match or exceed Anthropic salary and vesting, and that employees can reach him by direct message. He frames the public format as deliberate: a private message, he argues, would pressure already stressed engineers.
The Brazilian host of the video under review, Lucas Montano, adds his own experience on top of the letter. He says he canceled Anthropic $200-per-month Max subscription and moved to the $20 tier, citing repeated platform outages. That is first-hand user experience, not evidence about internal culture, and the two claims should not be blended.
How Did Anthropic Employees Respond?
At least one Anthropic employee pushed back on the letter in public. In the exchange quoted in the video, a researcher replied that the author knew nothing about conditions inside the company and was posting for attention. The author apologized for misreading the person's role, then defended the public format again.
The exchange turned sharper from there. The researcher asked why the offer had to be a public declaration at all, and the author answered that employees he had spoken with were too stressed to receive private approaches. The researcher called that reasoning generous to a fault. The video host reads the friction itself as a signal that research and product teams inside Anthropic not see the same company, which is an inference, not a documented fact.
Treat the whole thread carefully. It establishes that a public disagreement exists and that at least one employee rejects the letter's premises. It does not establish anything about Anthropic-wide working conditions, uptime, or code quality.
Why Does the Letter Attack Dario Amodei?
The letter's sharpest language targets Dario Amodei, Anthropic CEO and co-founder. The video host summarizes the criticism as a contradiction: Amodei publicly predicts software engineering will change dramatically within months, while the host claims the Claude platform suffers daily outages and the company pays engineers salaries of around $500,000 a year.
Only parts of that framing are checkable. Amodei's public predictions about the future of programming are documented statements from Anthropic leadership. The outage frequency and salary figures come from the video host's account and second-hand impressions, so they belong in the evidence class of speaker claims. No public status page data or payroll record is cited in the video.
The fair summary is this: the letter argues engineers are being asked to ship under pressure they did not sign up for. Anthropic not publicly confirmed reliability problems at the scale the host describes, and the company's own product releases in 2026 continued at a regular pace.
Why Did ChatGPT Keep Mentioning Goblins?
The second story in the video is verifiable on OpenAI's side. In early 2026, users noticed ChatGPT, OpenAI's assistant, randomly inserting goblin metaphors into replies. OpenAI published an explanation: model behavior is shaped by many small training incentives, and the personality customization feature had unintentionally given high rewards to creature metaphors. The goblin pattern then spread through the model's outputs.
The lesson the host draws is a real one about model training. A single overly rewarded behavior can propagate widely during reinforcement learning, even when nobody intended it. OpenAI's own short answer confirmed exactly that mechanism, which makes this one of the clearer public post-mortems of a reward-model quirk.
The two stories are connected in the video only by mood, not by fact. One is an unverified portrait of a competitor's internal culture. The other is a vendor-documented training behavior with a published cause. An open letter to Anthropic says nothing about how OpenAI trains personalities, and vice versa.
Two Stories, Two Evidence Standards
Keeping the evidence classes separate is the only way to read this material honestly. The table below sorts each claim in the video by who actually stands behind it.
| Story | Claim | Source type |
|---|---|---|
| Open letter to Anthropic | Employees should leave; culture is hostile | Creator's public post |
| Anthropic reply | Researcher rejects the letter publicly | Quoted exchange in video |
| Claude reliability | Daily outages, $200 plan canceled | Speaker first-hand experience |
| Engineer pay | About $500,000 per year | Speaker's unverified claim |
| ChatGPT goblins | Personality training over-rewarded creature metaphors | OpenAI's published explanation |
The pattern is consistent: the OpenAI story closes with a vendor explanation, while the Anthropic story closes with competing public voices and no company statement. Anyone citing either should say which one they are holding.
FAQ
- Who wrote the open letter to Anthropic? The letter came from Theo Browne, a developer and content creator who runs the site t3.gg. He says he wrote it to reach Anthropic engineers directly and offers to help them find roles at other labs without losing unvested equity.
- Did Anthropic respond to the letter? Anthropic a company has not issued a formal response. In the exchange quoted in the video, one Anthropic researcher publicly disputed the letter's claims and accused the author of posting for attention, which the author denied.
- Why did ChatGPT suddenly talk about goblins in 2026? OpenAI explained that the personality customization training gave unusually high rewards to creature metaphors, so goblin references propagated through the model. The company described model behavior as the product of many small incentives.
- Is Claude actually unreliable based on this video? The video offers one user's experience of outages and a downgrade from a $200-per-month plan to a $20 plan. That is anecdotal first-hand experience, not a measured reliability record, and Anthropic not confirmed the claimed outage frequency.
From Video Arguments to Written Pages
This video works because it separates what a creator claims from what a vendor confirms, and that separation is exactly what written content does best. If you publish arguments like these as videos, the same evidence discipline deserves a durable written form. Creators in the Brazilian dev scene, from Gustavo Dev Doido and his Crazystack Typescript community to the Bootcamp do Dev Doido, sit on hours of recorded reasoning that search engines never see.
Skala Blog turns that recorded reasoning into an article: paste a YouTube URL, get a transcription, and generate a structured draft you can review before publishing. For a deeper curriculum path, Crazystack Typescript at crazystack.com.br pairs well with written study material.
The open letter to Anthropic spread because it was text anyone could quote. Your own best arguments can work the same way when they exist as pages, not only as playback.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits