AI Worth Using · August 31 · 48 min

The Adversarial Review System Every AI Coding Agent Needs with Wes McKinney at Kenn Software

The creator of pandas doesn't read his own code anymore. His review system does, and it trusts nothing. Wes McKinney, creator of pandas and founder of Kenn Software, has agents write his software. RoboRev is what keeps that honest: an adversarial pipeline where a different frontier model tears apart every commit about 90 seconds after it's written, in the background, before it ships. Wes walks us through the origin (Claude Code in one terminal pane, Codex reviewing it in the other, dozens of times a day, until he automated it), why agents claiming "production ready" broke his trust for good, and why he says skipping automated review in 2026 is driving without a seatbelt. Tom installs RoboRev live on a fresh project, mistakes included, to see exactly how fast the loop runs. KEY TAKEAWAYS 00:00 Introduction 03:58 An adversarial code review pipeline sends every generated line through a different model before it ships 07:16 Early coding agents would claim finished, production-ready work that was neither finished nor production-ready 12:44 Regaining a sense of agency after months of doubting whether software engineers still have a role to play 21:31 Vibes-based bug finding validity lands somewhere around 70% to 80% for the best coding agents 35:52 An elaborate automated workflow raises an honest question: is this clarity or a touch of AI psychosis? 37:33 Keeping code review batches to roughly 70 lines at a time turns out to catch the most bugs 51:07 Spending $100 in tokens to review a 10,000-line pull request might be the easy call RESOURCES MENTIONED Wes McKinney LinkedIn https://www.linkedin.com/in/wesmckinn/ Kenn Software website: https://kenn.io/ RoboRev: https://roborev.io Superpowers (spec-driven development framework): https://github.com/obra/superpowers Pandas (Python data analysis library): https://pandas.pydata.org/ See how people are actually working with AI at https://aiworthusing.com/builder-index #AICodeReview #BuildInPublic #AITools

0:00-48:33

Play in

Transcript file

transcript

No transcript — this publisher did not publish one.

show notes

The creator of pandas doesn't read his own code anymore. His review system does, and it trusts nothing.

Wes McKinney, creator of pandas and founder of Kenn Software, has agents write his software. RoboRev is what keeps that honest: an adversarial pipeline where a different frontier model tears apart every commit about 90 seconds after it's written, in the background, before it ships.

Wes walks us through the origin (Claude Code in one terminal pane, Codex reviewing it in the other, dozens of times a day, until he automated it), why agents claiming "production ready" broke his trust for good, and why he says skipping automated review in 2026 is driving without a seatbelt.

Tom installs RoboRev live on a fresh project, mistakes included, to see exactly how fast the loop runs.

KEY TAKEAWAYS

00:00 Introduction

03:58 An adversarial code review pipeline sends every generated line through a different model before it ships

07:16 Early coding agents would claim finished, production-ready work that was neither finished nor production-ready

12:44 Regaining a sense of agency after months of doubting whether software engineers still have a role to play

21:31 Vibes-based bug finding validity lands somewhere around 70% to 80% for the best coding agents

35:52 An elaborate automated workflow raises an honest question: is this clarity or a touch of AI psychosis?

37:33 Keeping code review batches to roughly 70 lines at a time turns out to catch the most bugs

51:07 Spending $100 in tokens to review a 10,000-line pull request might be the easy call

RESOURCES MENTIONED

Wes McKinney LinkedIn
https://www.linkedin.com/in/wesmckinn/

Kenn Software website:
https://kenn.io/

RoboRev:
https://roborev.io

Superpowers (spec-driven development framework):
https://github.com/obra/superpowers

Pandas (Python data analysis library):
https://pandas.pydata.org/

See how people are actually working with AI at https://aiworthusing.com/builder-index

#AICodeReview #BuildInPublic #AITools

links6