How we build
Specs are tests. We use AI agents to write most of our code, which makes the writing the cheap part. The expensive part is knowing the code does what it should, and that’s what our method is built around.
The method
Every piece of work starts as a spec: plain sentences that say what the software must do. Each sentence gets a check. If a machine can run it, it’s an automated test. If it can’t, a named person signs it off.
Then the work is cut into tickets small enough for one AI agent to finish and one person to review. A ticket isn’t done when the agent says so. It’s done when every check in the spec passes and a person has read the change.
We plan to publish what each ticket costs, in compute and in human review time, in a public build log.
Why it matters to you
When you replace a tool your business runs on, the risk isn’t whether new software can be written. It’s whether the new software does everything the old one did for you, including the things nobody ever wrote down.
So we write those things down as checks before anything is built. You see them and you correct them. Then you decide when to switch, once they pass.
Where people come in
Some things a machine can’t judge: how a screen feels, whether a report is useful, what your team will put up with. People sign those off. And a check only covers what someone thought to write, which is why the spec is reviewed with you and a person reviews every change.
Pricing
We’re designing our pricing so that our fee comes out of what you save. Before you commit to anything, you get the numbers in writing: what you pay today, everything it costs after, and what’s left over. We’ll publish the exact model before we take on our first replacement.
Today
What you pay for the tool today
After
What it costs afterHosting, moving over and our fee
What you stop payingThe part of today’s bill that goes away
This website was built the same way, from a written spec, and the build checks what the site says about itself. See the checks