new pattern 'follow-obvious-fix'

[emilybache]

@devill and I used this a couple of times yesterday. I would appreciate some feedback on my writeup of it.

[lexler]

Thanks, Emily! I like it, a little brother of offload-deterministic.

A couple of small suggestions:

When I read it at first, I thought for a second this was advice for the AI (meant to be put in AI instructions).
Could we make it more explicit that this is for the human? Or did I misunderstand and it is actually for the AI?
You mention not wasting tokens, so I assumed it was for the human. "Don't ask AI to fix it, just do what the tool
suggested"

Also, the Problem section contains part of the solution; might be worth separating them. I think we usually put
solution at the start of the pattern section.

Wonder what you think about an alternative name Take the Hint? :)

Regarding the connections - I'd keep:

• similar to offload-deterministic - strongest connection, same shape
• solves hallucinations - following the tool's suggestion avoids AI making things up, makes sense

I'd suggest dropping these (let me know if I'm missing something):

• uses feedback-loop - it uses the feedback loop pattern if it's for the AI, otherwise I think this is more human in
  the loop steering? The feedback loop pattern is about giving AI access to all information it needs to make the
  right decisions on its own
• similar to active-partner - active partner is about encouraging AI to push back, give honest feedback and
  engage in dialogue
• solves keeping-up - what's the connection here?

[devill]

The interesting thing is that what I took from the experience is quite different. The way the pattern is described indicates to me that we should be manually pursuing the fix when this isn't what we were doing. Indeed we did the opposite: we offloaded that to the agent.

I created another pull request for my interpretation of what happened:
#22

@emilybache if you agree with my version I'm happy to add you as a co author to that one.

[emilybache]

Thanks for all your comments! I like the name suggestion 'take the hint'. Overall though I think we should go with @devill 's version of this instead.