no human -- code review

^ no human code review

Backlinks

blog post - OpenAI harness engineering 2026-02

Humans may review pull requests, but aren’t required to. Over time, we’ve pushed almost all review effort towards being handled agent-to-agent.
...
no human -- code review

see in context
free recall - Open AI harness blog post - 2026-02

First, and perhaps most importantly, they adopted a principal of no human -- coding, which has deep implications to their approach. They didn't say this explicitly, but it's not just the coding that they are attempting to remove humans from. They're trying to automate everything around the specification, development, and verification of an app. So, yeah, no human -- code review and no human -- app verification, too. They are even trying to reduce the amount of app specification that humans must do, too.

see in context

For code review I believe they have multiple agents attempting to review the code? I cannot recall whether they attempted to no human -- code review yet. But the before and after videos I mentioned would definitely reduce the amount of time a user would need to review the PR, if they do.

see in context
review - Open AI harness blog post - 2026-02-22
- I was unclear on what "enforceability" meant. I now think it means being able to run the gamut of validation scripts that are applied at every SDLC building block
  - These validation scripts run as continuous background checks
  - Could be similar to behavioral guardrails?
- They put a lot of effort into automating their code review, including it basically having its own ralph wiggum loop.
- It's not just no human -- coding, no human -- code review, and no human -- app verification. Within things like coding, it's about not needing to give any feedback. You shouldn't have to tell it "hey, you forgot a unit test... please add and TDD". That's far too involved, even if you aren't directly writing the code. Instead you need to try force that into the automated process. This feels like Kent Beck thought about what he did and distilled principles and practices into the Extreme Programming book, but these guys are doing it for agents, instead
- The onboarding a new teammate lens to agent context engineering. A Moc framing
see in context
tweet - design principles needed for AI coding

Unless you can afford to auto-merge PRs without a human review (and regulatory controls). But that requires a massively different level of effort - auto triaging the pull requests and a functioning suite of tests (fast & non-flaky with >70% coverage).

Triaging PRs is non-trivial for a complex codebase. A good classification system would look at hotspots, blast radius, architectural complexity etc., all of which are hard to measure and we don't have good tools for these.
...
no human -- code review auto-merging PRs

see in context