🔗 Impatient Programming: an invite for YAGNI listeners. This is a free guide to how Stephen’s team at Zar builds software with coding agents. It reworks Extreme Programming’s rules for working with agents and includes case studies, essays and tools. Nothing is for sale; signing up gives you access to everything.
Stephen Margheim leads product engineering at Zar, a stablecoin neobank for developing countries. His first big push with AI agents left a mess he had to read through and fix file by file. Now he runs ten sessions a day, reads maybe 10% of what ships, and feels more confident than before. He explains what changed: if code is cheap, keeping the codebase clean is worth more than ever. He and Matt talk about all-or-nothing changes, why he’d rather have a “software foreman” than a software factory, how he rewrote all of Zar’s money-movement logic in 30 to 50 small releases, and why quality checks should set floors, not targets.
Mentioned in this episode
Code Is Cheap Now. That’s Why It Matters.: The essay behind the episode. Because agents learn from the code around them, a messy codebase spreads mess into every session your team starts. Making it consistent is now cheap enough to do in one pull request, the way a team runs Prettier across every file at once.
I Don’t Need a Software Factory. I Need a Software Foreman.: Stephen’s essay on why the real value is agents paying attention around the clock, not generating code. His foreman agent’s system prompt isn’t “find and fix bugs.” It’s “notice anything worth noticing.” Layers of filtering then decide what’s urgent, what needs a closer look and what’s ready for a fix. That way you get short, clear flags instead of a stack of automated PRs.
Rubber-band AI: Matt’s Mario Kart theory that people who never learned the old way may slingshot ahead. Stephen’s reply is that they may also be falling behind without realizing it.
exhale: A Rails gem built around Stephen’s “inhale and exhale” rhythm, where agents explore freely and then pull the code back together. It fails a pull request when it finds duplicated code that hasn’t been listed as intentional, along with the reason, in a “Contract” file. That keeps agents’ spread-out exploration from piling up.
ACID: Stephen borrows the database idea for every code change. A change is “zero or a hundred,” so if a pattern is better, you make it true everywhere in that PR. Debates happen afterward, based on real use, and his team is on version 3 of its runtime constants system that way.
Putting an Agent in an Orb: A callback to the Thorsten episode, where an agent checks the browser console every day. It’s an example of Matt’s point that a lot of this watching takes patience more than brilliance.
Goodhart’s law: The idea behind floors versus targets. Chasing numbers leads to lots of small, over-tuned pieces. A minimum bar, like “below 80 on evals means no,” keeps a system simple, with fewer parts that each do more.
Thank you to this season’s sponsor Laravel Cloud. Laravel Cloud is expanding beyond PHP to host your Rails, Node, or Django apps. For early access to the best place to host your app, email haley@laravel.com and mention YAGNI
Subscribe
Listen to YAGNI using one of many popular podcasting apps or directories.
S3E6