Non-Tech Staff Promise 2-Day AI Timelines While Evals Struggle to Catch Up

kmeanskaran · x · 2026-09-25

Engineer kmeanskaran describes a widespread pain point: product and sales people commit features on timelines they believe AI can deliver in days—assuming development works like generating an image in ChatGPT. AI does speed things up, but pushing new features in short windows is unhealthy.

His proposed fix: lean on evals and observability, using an LLM-as-a-judge for verification from both technical and business perspectives to eventually balance latency vs accuracy. He also mentions Jev as an option, though unsure how mature it is.

Related event: Engineers suffer as non-technical colleagues promise AI-speed delivery(2 posts)→

Original post →

More from coding & agent

coding & agent channel →