How do you know your coding agent actually helps?

Boris Paskalev CEO & founder of LogicStar.ai

Time

16:00–16:30

Room

Studio 4

Times and rooms can change at any time, also during the conference. This page always shows the current schedule, check it on the day.

Agents now write most PRs in many teams. Often another agent reviews them. Yet almost nothing merges without a human, because nobody can show where the agents are safe to trust. Writing code got automated. Trusting it didn't.

Public benchmarks won't tell you. On real applications, agents resolve 38.6% fewer issues than on popular libraries (SWA-Bench, ICML 2025). Your git history is a better benchmark. Every revert, hotfix and follow-up fix points back to a PR, the agent that wrote it and the review that missed it.

This talk shows how to measure coding and review agents on what actually shipped, with data from real teams. 1 in 5 PRs that a review agent graded "perfect" still introduced a bug. On 78% of PRs, human review changed nothing. We trace one escaped bug end to end, go through the four failure modes behind most review misses, and give you a recipe to run on one repo next week.

★ Spotlight Talk by