Will AI replace programmers? What the people building AI coding tools say
September 29, 2026 · 25 min read
The Cursor founders, Dario Amodei, Sam Altman and ThePrimeagen each describe how programming is changing as models write more code, and they disagree on what remains uniquely human. Their views matter for anyone deciding whether and how to learn to code now.
The rules, numbers and steps worth keeping in front of you.
- Nobody in these sources predicts programmers disappear soon; they disagree on what stays human.
- Dario Amodei: programming changes fastest because models can write, run and check code in a closed loop.
- Amodei: human work shifts to design, architecture and user experience, until AI does those too.
- Amodei: a software benchmark went from about 3% to about 50% in a year.
- Cursor founders: engineering is countless micro-decisions; handing them off loses control.
- Cursor founders: agents are not yet very useful for most tasks, and models are bad at finding bugs.
- Cursor founders: review and verification grow as AI writes more code.
- Cursor founders: benchmarks are contaminated and miss messy, half-specified real work.
- Sam Altman: ask what share of tasks AI can do, not how many jobs it replaces.
- ThePrimeagen: AI is great for predictable work and weak in cutting-edge or poorly documented areas.
- ThePrimeagen: use AI as a co-pilot, not an autopilot; reading code is a different skill from writing it.
- ThePrimeagen: he learns most when a project reaches 5,000 to 10,000 lines.
- Only about 2.7% of surveyed developers highly trust AI output.
- Caution: three of the four sources sell AI coding tools or models.
This is the short form. The full insight has every expert's reasoning, their own words and the sources.