24 Comments
User's avatar
Karen Spinner's avatar

Great resource! I know from personal experience how important it is to do the product thinking up front and give coding agents well-documented standards…and what happens when you don’t! 😆

I did this for my most recent project and have experienced so much less chaos…

Karo (Product with Attitude)'s avatar

Thanks for sharing that, really appreciate you bringing your own experience into this discussion!

Alejandro Baena's avatar

same approach I started. a command file that clearly guide the llm, only upload the custom input for each new work. so value your post, I will adjust and improve my strategy with it. thanks again @Karo

Karo (Product with Attitude)'s avatar

That's great! Thank you for reading Alejandro! 🤗

Robert Richman's avatar

Wow. Impressive.

Karo (Product with Attitude)'s avatar

Thank you for reading Robert!

ToxSec's avatar

“Agent files aren’t static, they grow with your project.

This workflow gives you a rock-solid starting framework,

but the real power comes from how you evolve it. “

I feel like this is something that is often overlooked. Great post thanks :)

Sharyph's avatar

This is a super interesting take. I have noticed the same thing, where the most robust technical guidelines for agents often miss the product-level context a beginner (or even an experienced engineer without a PM background) needs to avoid building something that collapses under its own weight later.

Data Frank's avatar

The way you described building systems at the kitchen table while someone asks where their shoes are made me smile.

I’m doing the same inside My Accountability Partner, turning chaos into repeatable wins, one tiny rule at a time.

What’s the smallest system you rely on daily?

Karo (Product with Attitude)'s avatar

Thank you for reading Data Frank. Not sure which one is the smallest :)

Data Frank's avatar

Oh!

Alright

Anthony's avatar

You’re so right! agent instructions really are the new dev secret sauce. Thanks for sharing your findings.

I also love Cursor’s “generate rules” feature to bootstrap a project-specific rule set and refine it as you go.

Two additional issues I keep running into with agent rules:

• Finding the right level of instruction. Too many rules and the agent starts skipping some, while we assume it will follow all of them. Do you have any mental limit or heuristic for that?

• I rely heavily on project documentation, also stored as .md files. The boundary can get blurry between what belongs in documentation vs. what belongs in instructions (or conditional instructions — e.g., only when touching the database, or UI features…).

Anyway, thanks a lot for your work, Karo!

And I’d be more than happy to join the next community session.

Karo (Product with Attitude)'s avatar

This is such a thoughtful comment, thank you for taking the time to write it! 🤗 Looking forward to seeing you there!

Suhrab Khan's avatar

Really appreciate how this system makes AI rules files accessible and practical for beginners, such a thoughtful approach!

Karo (Product with Attitude)'s avatar

Thank you for reading Suhrab!

Sufyan Maan, M.Eng's avatar

thanks for sharing, Karo

Karo (Product with Attitude)'s avatar

Thank you for reading Sufyan 🤗

Raghav Mehra's avatar

Amazing resource this, Karo! How this Agent Rules Generator prevents shipping features that break the billing, onboarding, and other flows we forget to think about while building is a big value-add.

Karo (Product with Attitude)'s avatar

It really is. Thank you for reading Raghav!

Ileana's avatar

Thank you for sharing this, Karo! Amazing ✨

Karo (Product with Attitude)'s avatar

My pleasure, thank you for reading!

Rakia Ben Sassi's avatar

Many thanks for the shout-out, Karo!

Joel Salinas's avatar

Thanks for the mention :)