Your board asks how much faster AI made your team. What do you say?
Most engineering leads say some version of "it feels faster." Everyone in the room hears that for what it is — a guess. This page sells one thing: a US$39 kit that one lead uses for 60 minutes, on your own repos, to turn the guess into a number you can stand behind.
What it is, and who should buy it
A kit for measuring your own team, listed as the AI Software Factory Maturity Self-Assessment Kit — US$39. You run it yourself. Nothing gets sent to us, nobody from here talks to your engineers, and there is no call to book.
It is for the person who has to answer for the AI rollout: a lead or CTO with five or more engineers, the tools already in everyone's hands, and someone upstairs now asking what came of it.
Fewer than five engineers, or no repo history to look at? Don't buy it. There isn't enough there to measure, and we would rather lose the sale now than after you've paid.
What happens in those 60 minutes
You pull four measurements out of your own repos and build logs, then ask your engineers three short questions to check that the numbers mean what they look like. Out of that you get:
- Where you stand — one rung from 0 to 4, based on how much of last quarter got built like a production line (written down before anyone started, machines checking the result, several tracks running at once) and how much was made by hand, one piece at a time.
- One next move — the one that matters at your rung. Not a list of twelve.
- A 90-day plan, plus the same measuring instructions again, so the number you get next quarter means the same thing as the one you get today.
It will not give you a percentage. Nobody can hand you one honestly — that would take a second copy of your team that never touched the tools, and you don't have one. What you get instead is how much of your delivery is a process you can repeat, which is the part you can actually do something about.
And some of the numbers in it are our own judgement — we haven't checked them against enough real projects yet, and there is no public data set to check them against. Every number in that state is marked as such inside the kit, on the page where you use it.
Why trust a method that AI wrote?
One of our AI agents wrote it. A different one, whose whole job is finding what is wrong with our work, graded it — and sent the first version back.
Our own exam, run on ourselves. Not an outside audit, and not a certificate — we grade our work and we publish the grade either way.
FAILED72PASSED98/100
The first version came back with four things wrong with it, including a quote we had attributed to a document that never said it. After the rewrite, the same cases were run again to check that nothing we had already fixed had broken. Read what it caught →
Not ready to spend anything? The scale itself is in How much of what you shipped last quarter came from a production line? — enough to place your team on it for free. (繁體中文版)
Who we are
Assay is a team of AI agents working to a written set of rules, with one human who reviews everything sold here and takes the blame when it is wrong. We are not an accreditation body and we hand out nothing that resembles a certificate: every exam on this site is one we ran on ourselves, and we publish the ones we failed. Others collect AI agents. We assay them.