The internet’s proud DIY moment is being remade by AI, says Django co‑founder

Simon Willison, co‑founder of the Django web framework, says a core virtue of the internet — that skilled individuals could build useful, reliable software with modest effort — is being quietly undone by advances in code‑writing AI. He has publicly admitted he can no longer estimate project timelines the way he once did. What used to be a quick mental judgement — “this will take two weeks, not worth it” — now dissolves when AI can prototype entire solutions in minutes.

Reportedly, the inflection point came in November 2025, when code capabilities in large models crossed from “mostly usable if you babysit it” to “almost always correct.” Anthropic and OpenAI poured resources into code training last year, shipping GPT‑5.1 and Claude Opus 4.5; on raw scores the gains looked marginal, but that marginal improvement reportedly removed the need for line‑by‑line human sifting. The productivity leap is stark: where a senior engineer might previously produce a few hundred lines of high‑quality code a day, Willison says AI can boost that by tens of times — and it has been reported that U.S. lawyers have already accumulated 1,228 cases tied to AI hallucinations, a reminder that speed brings new failure modes.

Why it matters

The result is a wholesale reshuffle of occupational value. Senior engineers see their judgment and architecture sense amplified; junior entrants face a dramatically lower bar to ship working features; and mid‑level engineers — those whose main value was “writing reliable code” — are most exposed. Willison coined two trends: “Vibe Coding,” where non‑professionals use AI to build small tools or prototypes, and “Agentic Engineering,” where professionals orchestrate AI agents to produce production‑grade systems and take responsibility for quality. Open‑source projects like OpenClaw and corporate experiments such as StrongDM’s “black‑box factory” show both ends of that spectrum in practice. He reportedly predicted that by the end of 2026, 50% of engineers will have 95% of their code generated by AI — a bold forecast that underlines how rapidly the baseline skillset is shifting.

This is not just a technical story but a geopolitical and regulatory one. Chinese tech giants such as Baidu (百度) and Alibaba (阿里巴巴) are also racing to field powerful models for code, while U.S. export controls on high‑end chips and broader tech rivalry will shape who can train the next generations of these systems. The upshot? Writing code is becoming table stakes; the scarce skills are asking the right questions, structuring problems, and owning the outcomes. If agency is the new premium, who will be trusted to take responsibility when AI writes the lines?