[vova_hn2]

On the one hand it is a cool and creative idea, on the other it feels to much like being handed an assignment by a teacher at school, or something like this.

And I am not comfortable putting a clanker in a position of authority over me.

Also, it might decrease LLM output quality, because in the process of code generation it will be lacking parts of code from it's context.

Although, it can be mitigated by first generating all the code and then removing some of it and replacing with HUMANTODO.

In this case it would also be very interesting to compare the LLM-generated solution with human-generated and then pick the one that you like more.

IDK, now I'm actually curious to try

[bodash]

I can see how some high-paced teams might see HUMANTODO as speed bumpers in their environments. To that, I say this will be a conscious choice, and a trade-off teams are willing to experiement to see if the intentional slow-down will improve quality over time.

But yeah, give HUMANTODO a go, and I might add examples (blog/videos/etc) of how this actually works and integrates within real life projects.