What happened
Copilot had been a technical preview with more than a million developers. General availability sold the product at $10/month and, GitHub said, was already writing nearly 40% of code in files where it was enabled for popular languages. The underlying model was OpenAI Codex, a GPT-3 descendant fine-tuned on public code. Peng, Kalliamvakou, Cihon, and Demirer later reported a randomized experiment (n=95) in which treated developers finished a standardized task 55.8% faster (95% CI 21–89%). That study measures a narrow task, not firm-level output, and later field studies of AI coding tools have been mixed. The event is still the product launch that made inline AI coding a default part of professional software engineering, not a benchmark score and not a user-count trophy. Update (2026-09-06): Research acceleration: The view inside OpenAI ↗. Inside OpenAI, coding agents are reshaping AI research. Explore early data on agent usage, experiment velocity, task complexity, and research acceleration. Update (2026-09-21): AI coding has made CI a bottleneck, so we reworked ours to keep up. Article URL: https://linear.app/now/ci-bottleneck-reworked Comments URL: https://news.ycombinator.com/item?id=49792067 Points: 14 # Comments: 4