Analysis · 30 September 2026 · 6 min read

Dario Amodei's software engineer prediction

In January Amodei said AI would do most software engineering within 12 months. Eight months in: benchmarks, job postings and hiring data, checked.

At Davos on 21 January 2026, Anthropic CEO Dario Amodei said AI models could be doing "most, maybe all" of what software engineers do, end to end, within six to twelve months. The window runs out in January 2027. Eight months in, the models have moved a long way toward that line. Software jobs have not disappeared: postings are up since early 2025, but the gains are going to senior roles while hiring of young workers has fallen.

What he actually said

The prediction, made in an interview with The Economist at the World Economic Forum, was about what models can do, not about unemployment. He described engineers at Anthropic who say: "I don't write any code anymore. I just let the model write the code, I edit it." He also said he was not certain how fast the change would come. Coverage of the interview has the full context.

So there are two separate questions: are the models close to doing the job, and are the jobs going away.

Are the models close?

On the benchmark closest to day to day engineering work, yes, and quickly. Terminal-Bench 4.0 tests long, multi step work in a real terminal.

Figures from Anthropic's Opus 5.5 and Sonnet 5.5 launch pages.

Anthropic also reports that Opus 5.5 audited and fixed a 200,000 line codebase in under three hours, a job that took Opus 5 more than twenty. Our post on that release covers the detail.

But "most, maybe all" is not there yet. On FrontierCode, a harder coding benchmark, the best Claude model scores 54.4 percent, so it still fails close to half of those tasks. The deliverables are also not trusted without review. Amodei's own description, engineers who edit what the model writes, is still how the work is done.

Are the jobs going away?

Not overall. Indeed's Hiring Lab reported in July 2026 that US software development job postings had grown almost 15 percent since Claude Code launched in February 2025, while overall postings fell 7 percent. They are still about 27.5 percent below February 2020.

The detail matters more than the total. 71 percent of the increase between May 2025 and May 2026 came from senior roles, and 37 percent from jobs with AI in the title.

At the bottom of the ladder the picture reverses. The Stanford Digital Economy Lab's August 2026 update found employment of workers aged 22 to 25 in AI exposed occupations 19 percent below where it would have been had it kept pace with less exposed peers. Our post on entry level jobs explains why the youngest workers are hit first.

Why the job split matters

In Anthropic's own usage data, the difference between two software occupations is large:

Observed exposure is the share of an occupation's tasks showing up in automated, work related use, from Anthropic's March 2026 labor market study. Task times are from Anthropic's June 2026 Economic Index release; blanks mean the release has no figure for that occupation.

Writing code to a specification, the core of the programmer role, is the part models already do. Designing systems, deciding what to build and owning the result are not. That is the same split the Indeed data shows: senior demand up, junior hiring down.

In our index, which combines published research rather than a single usage measure: Computer Programmers (59) and Software Developers (47).

Where this leaves the prediction

With about four months left, the capability half of the claim is closer than most people expected in January, and the jobs half has not happened. If the prediction is judged by his own words, models doing most of the work while engineers edit, it is on track for many teams. If it is judged by headlines about software engineers being replaced, it is not.

For a similar prediction about lawyers, accountants and marketers, see Mustafa Suleyman's 18 month forecast.

Where your own job sits

If you write software, the useful question is how much of your week is writing code to a specification, and how much is deciding what gets built. The assessment scores that for your role in about two minutes. It is free and needs no account.

Figures come from the AI Job Risk Index and were current when this was published. Scores change as new research is ingested, so the index is always the live version. See how scoring works. Informational guidance based on published research, not professional career or financial advice.