Simon Willison, co-creator of Django and the person who coined 'prompt injection,' says November 2025 was the moment AI coding agents stopped being a party trick. GPT-5.2 and Claude Opus 4.5 crossed a threshold where agentic coding actually works at production scale. His prediction: 50% of engineers will be writing 95% AI-generated code by end of 2026. He is already there, writing most of his code from his phone.
The conversation earns its runtime because Willison does not just describe outcomes, he describes mechanics. He breaks down three agentic engineering patterns he uses daily: red/green TDD to constrain AI output, project templates to establish context fast, and deliberate 'hoarding' of skills the AI cannot verify on its own. He also names the 'dark factory' pattern, where AI writes, runs, and QA-tests its own code with no human in the loop. That is not a future scenario. It is a current architectural pattern. Mid-career engineers, not juniors, are identified as the group most exposed, because they carry assumptions about workflow that do not survive contact with this environment.
The security section alone justifies a full listen. Willison coined 'prompt injection' and considers it fundamentally unsolved. He defines a 'lethal trifecta' of conditions that he believes will produce an AI equivalent of the Challenger disaster, a high-profile, preventable failure caused by normalized deviance. He also flags OpenClaw as a security risk the industry is actively ignoring. The pelican-on-a-bicycle benchmark, his informal test for model quality, is a small detail that reveals how he actually evaluates tools, and it tells you more about his methodology than any benchmark leaderboard would.
[WATCH ON YOUTUBE →]