Skip to content
Blog

Codify the conversation

Codify the conversation
AuthorMary Hurd
Date

October 29, 2026

We’ve been building A LOT of agents with our partners lately, helping teams build skills and agents around real work. Our #1 hot tip: ✨ codify the conversation ✨

It’s a quick way to discover how much of your work you’ve never explained to anyone else. You know which sources you trust, what makes an answer useful, where things live, and when something needs input. Most of these decisions happen without you thinking about them. A skill gets them out of your head.

Pick a task you do often and work through it with AI. Explain the steps, share the context it needs, review what it produces, and tell it what you’d change and why. Keep going until you’re happy with the result. Then ask AI to turn the whole conversation into a skill. Literally type, “turn this conversation into a skill.” The next time that task comes up, AI already knows how you like it done.

You end up capturing two things: an example of what good looks like, and the messy history of how you got there. The corrections, feedback, and logic you give along the way are what make the result useful, so they’re worth keeping.

It also helps to treat skills and agents like mini products. Start by understanding the problem: what do you actually need help with, and what does a useful result look like? Then test it on real work. A writing skill might nail your emails and still need more guidance for Slack. A knowledge assistant might pull an answer from an outdated doc or database.

Each time something goes wrong, tell the AI what happened, work through the fix together, and ask it to update its instructions so the learning carries forward. You don’t need to edit anything manually.

Creating agents adds a few more decisions, like when they run, what they can access and change, and when they check in with you. The first version gives you something to learn from, and using it shows you what to improve next.

I find this part of building with AI super effective. It makes us put into words how we actually work, then turns that into something we can use.

It also takes time to build, test, notice what breaks, and improve. That feels impossible on back to back days when you still have your actual job to do. But what we hear most from the people we work with is that they need time to build, and making room for building is how tools become useful parts of the team.

✨ Codify the conversation! Curious what you discover about your own work while teaching an agent to do it.