Reading Time: 7 minutes

Hey Prompt Lover,

After last week's newsletter on Sam Altman, the replies came in fast.

Most of them said some version of the same thing. "I appreciate you covering OpenAI but don't forget Anthropic. Cover Dario too. Cover them all."

Fair. That was always the plan.

This is not a series about OpenAI. It is a series about the gap between what AI leaders say publicly and what actually happens. That gap exists at every company building this technology. It deserves to be tracked at every company building this technology.

Today it is Dario Amodei's turn. The co-founder and CEO of Anthropic. The company that makes Claude. The man who has built a reputation as the thoughtful, safety-first alternative to the move-fast culture at OpenAI.

The record is more complicated than the reputation.

Elon Musk is coming. Other CEOs are coming. Nobody gets a pass in this series.

Most AI agents can run a task. The problem is everything around it: setup, memory, context, cost, and figuring out what actually happened.

SureThing turns useful AI skills into autonomous agents with business context, persistent memory, cost-aware model selection, and a live dashboard. Paste a link, assign the work, and your agent reports back like a human teammate: what it did, what it cost, what needs your decision, and what happens next.

Built for founders, operators, and marketers who want AI to ship work, not become another tool to babysit.

Create Your Agent

What he said — March 10, 2025, at a Council on Foreign Relations event:

"I think we'll be there in three to six months, where AI is writing 90 percent of the code. And then in twelve months, we may be in a world where AI is writing essentially all of the code."

Not ten years. Not five. Three to six months. From the CEO of one of the most credible AI companies on the planet.

What happened:

It's been about a year now. 84 percent of developers use or plan to use AI tools and 51 percent use them daily. That's widespread adoption. It is not near-total replacement. GitHub data puts AI-assisted code at around 46 percent of files where it's enabled, with suggestion acceptance around 30 percent in enterprise settings.

When challenged on this six months later, Amodei said the prediction had come true at Anthropic specifically. Researchers who looked carefully at what he actually said found that if you include code which was at all useful including things like scripts which only get run once, the fraction written by AI at Anthropic is higher, probably closer to 90 percent than 50 percent, but this is hard to measure and exactly what you include might make a big difference. The prediction was not what most people interpret it as when he said "this is absolutely true."

He made a sweeping industry prediction. When it didn't land industry-wide, he redefined it to apply specifically to his own company. Then researchers looked at his own company and found even that was contested.

What he said — January 2025, Davos:

"I don't know exactly when it'll come, I don't know if it'll be 2027. I think it's plausible it could be longer than that. I don't think it will be a whole bunch longer than that when AI systems are better than humans at almost everything. Better than almost all humans at almost everything. And then eventually better than all humans at everything, even robotics."

Anthropic's official position, submitted to the White House Office of Science and Technology Policy in March 2025, states they expect powerful AI systems to emerge in late 2026 or early 2027 — with intellectual capabilities matching or exceeding that of Nobel Prize winners across most disciplines including biology, computer science, mathematics and engineering.

What happened:

We are now in May 2026. Eight months from the lower end of that window. The jury is out on whether this one ages well or doesn't. But the boldness of the claim deserves to be noted now, while there is still time to check it.

A Nobel Prize winner in every discipline. By early 2027. From the CEO who also says he is deeply uncomfortable with the power a handful of people have over this technology.

What he said — 2025, repeated in multiple forums:

Amodei warned AI would destroy half of all white-collar jobs. He predicted that humans would be unable to adapt to AI development's rapid pace and this would trigger an "unusually painful" short-term shock in the labour market. He said tackling this problem would "require government intervention" such as "progressive taxation" targeting AI firms specifically.

What happened:

Nvidia CEO Jensen Huang publicly said he disagrees with almost everything Dario Amodei says and pushed back specifically on the idea that AI could soon replace half of all entry-level office roles.

AI was given as a reason for nearly 55,000 layoffs in the US in 2025. A study by MIT found AI can already do the job of 11.7 percent of the US labour market.

The displacement is real. The "half of all white-collar jobs" number is one that Amodei has not been asked to substantiate publicly in any rigorous way. It spreads. It shapes policy conversations. Nobody checks the working.

What he said — 2025:

Dario Amodei says he is "deeply uncomfortable" with unelected tech elites shaping AI and is uneasy that a few unelected tech leaders — including himself — hold the power to shape humanity's AI future.

What happened:

Anthropic's revenue grew from zero to $100 million in 2023, from $100 million to $1 billion in 2024, and to well above $4 billion annualised in 2025. Its eight and nine figure deals tripled in 2025 versus 2024. Its business customers spent five times more on average.

The man deeply uncomfortable with the power he holds has built one of the fastest-growing technology companies in history. Which he continues to scale at maximum speed while being uncomfortable about it.

The discomfort has not slowed the building. It has made excellent press.

Dario Amodei is genuinely one of the more thoughtful people in AI. His essay Machines of Loving Grace is worth reading. Anthropic's safety research is serious work done by serious people. Claude is a genuinely good product and the people building it care about doing it well.

None of that makes the predictions above more accurate than they turned out to be.

The point of this series is not to find villains. It is to notice that the people with the most information about how this technology works are also frequently wrong about it in public, without consequence, and without anyone comparing what they said against what happened.

That is worth tracking. For all of them. Without exception.

Next week — Elon Musk. The receipts on that one are extensive.

Reply and tell me which one surprised you most.

— Prompt Guy

This newsletter goes out to over 95,000 people every week. If you have a product or AI tool you want introduced to this audience, reply with "sponsorship" and I will send you the details.