aiqbench analysis
This talk pushes back on hype about AI writing "essentially all" code, citing estimates of roughly 25-50% depending on how code is counted. Its more useful claim concerns the labour pipeline: if juniors rely on AI tools before building foundational skills, organisations may lose the people needed to oversee those systems later. Readers should watch whether the workforce and training effects it describes show up in measured data, not just in anecdotes.
Presentation: The Ladder is Missing Rungs: Engineering Progression When AI Ate the Middle

Alasdair Allan discusses empirical research from METR and Anthropic on AI coding adoption. He explains the "supervision paradox," where reliance on AI tools disrupts junior skill development, risks institutional context loss, and threatens the long-term engineering pipeline needed to oversee autonomous systems effectively. By Alasdair Allan
Alasdair Allan: Not 5 years ago, but 12 months ago now, almost to the day in fact, someone else stood up at an entirely different conference in New York, and during their talk they said this. They said that within 12 months AI will write essentially all of the code. I'd expect some of you in here today still write code, or at least some of you run teams that still write code by hand, like savages. The person who said this was Dario Amodei, the co-inventor of RLHF, the training technique behind virtually every modern AI, and the co-founder of Anthropic. He said that within 3 to 6 months, AI will be writing 90% of all the code, and within 12 months, virtually entirely all of it. At the six-month mark Amodei claimed vindication saying that 90% figure was absolutely true, at least within Anthropic. Twelve months on, almost to the day, and that's not exactly what happened, is it? A detailed analysis by Redwood Research concluded that the 90% number he quoted was misleading. If you count only code committed to repositories, it was closer to 50%, which, at least to me, is still a pretty shocking number.…
Excerpt shown under the publisher's RSS terms · 200 words · full text and figures are on the publisher's site
Related coverage
Beyond Executable Models: The Pufibara Agent Harness and the Modelica Agent Workflow Benchmark for Physical System Modeling
arXiv · Aug 27, 2026
XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMs
arXiv · Aug 6, 2026
TIDE: Temporal Incremental Draft Engine for Self-Improving LLM Inference
arXiv · Sep 26, 2026

Presentation: Beyond Prompting: Context Engineering for Production-Grade AI
InfoQ · Sep 2, 2026