We gave an agent last quarter's results, the targets and our backlog, and asked it for a plan. It produced a very tidy plan that spread effort evenly across everything, which is the one thing a plan should not do.
What worked: asking it to argue against each initiative. The objections were sharper than the plan. We kept the agent as the critic and wrote the plan ourselves.
Critic not author is where we ended up too. It is relentless and has no favourite project.
We tried it too. The agent's plan was fine and nobody believed it, because nobody had argued about it. The arguing is the planning.
Useful for one thing: listing what we said we would do last quarter and did not. It has no feelings about that list.
The honest version is the only version worth reading. Thank you for not writing the other one.
Product marketer, just joined. We do the same for launch plans. 'List the three ways this launch fails' is my most-used prompt.
Add 'which of these have we already seen happen' and it gets very uncomfortable, in a good way.