OpenAI's Lukasz Kaiser: even we can't pinpoint what caused the Christmas coding-agent jump

a_karvonen · x · 2026-09-04

OpenAI researcher Lukasz Kaiser says the sudden coding-agent breakthrough around last Christmas is hard to attribute even internally: not a big pre-training shift, but likely the harness changes, post-training tweaks, and new models stacking up until continuous improvements crossed a threshold. A rare insider take on the non-linear emergence of agent capabilities.

Original post →

More from coding & agent

coding & agent channel →