[lonelyasacloud]

> ... and yet project velocity does not go much faster

1) The models like us have finite context windows and intelligence, even with good engineering practices system complexity will eventually slow them down.

2) At the moment at least, the code still needs to be reviewed and signed off by us and reading someone else's code is usually harder than writing it.

[flurdy]

Are people still reading PRs in detail manually?

I am after the automated PR agents have all passed a PR I tend to let Claude Code and Codex give me a summary, with an MCP skill to read the requirement story. I trust their ability to catch edge cases and typos more than me. I just check the general structure of the PR

[flurdy]

I may use my own skill to automate this... https://github.com/flurdy/agent-skills/blob/main/skills/revi...

[lonelyasacloud]

Anthropic, OpenAI, Google et. al. have EULAs and the best lawyers money can buy ready to argue that any damage done by publicly releasing bad or malicious code produced or reviewed with their systems is the developers responsibility for not checking properly.

> I trust their ability to catch edge cases and typos more than me.

Given the vendors EULAs etc, if poop really hits the fan with released code, then how is that likely to sound if the lawyers get involved?

> Are people still reading PRs in detail manually?

Ultimately it all depends on circumstance and appetite for risk, but yes many/most places still manually checking releases.