A markdown file is an employee that will do the job perfectly every single time and can be reused infinitely via cron jobs
Garry describes how agentic AI systems work: you create skill files (markdown + code + tests) that encode perfect execution of business processes. Once built, these can run indefinitely, enabling small teams to achieve massive scale. ✦ AI generated
Garry Tan · a16z Podcast · 2026-08-12 · original ↗
starts at this moment · 24:17
“Talk about how YC is organizing around loops and how you're advising your companies to.”
I think we're just seeing like across the board, a markdown file is an employee. And it's an employee that will do the job perfectly every single time and it will do it like as many times as you want. And then, you know, at that point all you're doing is like take any business process that you need in your company and you just do it once perfectly. And the first time you try to do it, it's going to be bad. But then at the end you can... turn that into a skill file that's perfect, actually. And anytime it screws up in like a future case, it's just a bug fix, and then it's there forever.
verbatim transcript · starts at 24:17
24:17want. >> Yeah. >> And then, you know, at that point all you're doing is like take any business process that you need in your company and you just do it once perfectly. And the first time you try to do it, it's going to be bad. >> Expensive. Yeah, yes, many iterations. >> Yeah. But then at the end you can you as long as you can tell very, very quickly and in in very
24:36simple ways, like, "Hey agent, fix this, fix that. Like this was wrong." Um, you know, the actual trace and the actual, uh, history out of, um, that agent, uh, working with you, >> Mhm. >> it'll actually turn that into a skill file that's perfect, actually. And anytime it screws up in like a future case, it's just a bug fix, and then it's there forever. >> And this loop you would apply to sales,
25:00to marketing, to customer support, to everything. >> Yeah, to everything. >> Really interesting. >> I mean, G stack was basically that for engineering, right? It's like there's a QA loop. I had to build something that works with browsers to go and cuz that's what I found like early with G stack, I found myself um, you know, automating how I thought about PM, and then uh, you know, I had a
25:21engineering skill that was like an engine manager that made sure that there was a unit test and end-to-end test coverage. >> Yep. >> Um, but then in the end, I found myself like uh, I automated all the other things so much that all I was doing at the end was black box testing to to verify. And it was like, "Oh, no, no, okay, like even that part." Like it's interesting
25:42like uh, building these agentic systems today, I think that's all you're doing is you're seeing where the bottlenecks are, and then, you know, almost every single time it's like, "How do I just instruct the agent to create a piece of software or a markdown file that blows away that roadblock." And then, when you take a step back and you've done that for the whole task, it's like, I mean, we we there are