Most founders pitch me an idea. The good ones already have a system, they just haven't noticed it yet. It's the spreadsheet someone updates every morning, the Slack thread where deals actually close, the manual review that quietly keeps quality high. That system is already creating value. The job isn't to invent something new. It's to turn what works into a product people pay to use.
Ideas are cheap, systems are proof
An idea is a guess about what might create value. A system is evidence that something already does. When you build from the idea, every decision is a bet. When you build from the system, most decisions are already made for you, by the people who've been running it by hand. Your customers have been beta-testing the workflow for months. You just haven't shipped the UI yet.
So before writing a line of code, I map the existing system:
- What's the input, and where does it come from today?
- What judgment call turns that input into something valuable?
- Who does it, how often, and what does it cost in hours?
- What would break if that person took a week off?
The last question usually finds the product. The thing that breaks is the thing worth automating first.
AI changes the economics, not the principle
What's new is that the judgment call in the middle, the part that used to require a person, can now run on a model. The expensive manual step becomes a function. But the principle is older than AI: ship the system that already works, then make it faster and cheaper than a human could.
You don't catch up to an exponential curve by planning. You catch up by shipping the thing in front of you and letting the next week compound.
What this looks like in practice
A two-to-three week sprint. Week one is discovery, we map the system and cut everything that isn't load-bearing. Week two is the build: the workflow, the model in the loop, the interface thin enough to ship. Week three we deploy and watch real users touch it. No half-built mess, no overbuilt platform waiting for users who never arrive.
If you've got a manual process eating your weeks, that's not a problem to complain about. That's a product brief. Let's ship it.
