Your customers are starting to point agents at your product. Most software wasn't built for that — and it shows the moment someone tries. We fix it.
No pitch. We spend a few hours with your product and send you what breaks.
For thirty years, software assumed one kind of user: a human with a screen and patience.
That user now has company. An agent has no screen, infinite patience for structure, and zero tolerance for ambiguity — and it can't guess what a greyed-out button means. There is no "mobile-ready" for this yet. There will be.
What we do
Most clients start with a teardown and end up somewhere below.
We take a product built for humans and make it work for agents too — designed tool surfaces instead of CRUD sprawl, errors a model can recover from, docs a model can read, and an auth layer that lets an agent act for a user without handing over the whole account.
Where AI would actually help you, and where it would waste your money. Every engagement ends in something that runs — a prototype, an eval harness, a fix list with evidence from your own traces. Never just a document.
Grounded in your actual content, not a generic model wearing your company name. It says "I don't know" instead of inventing an answer, hands off to a human with the full conversation attached, and gets better every month from real transcripts.
The readiness model
We score every product against the same nine dimensions, 0–4 each, with evidence. Most products in 2026 land between 6 and 12 out of 36.
Is there any programmatic way in — or is it a browser session and a CAPTCHA?
Operations shaped like tasks, or sixty endpoints shaped like your database?
Do the names mean what they say? status: 2 teaches a model nothing.
Does the error say which field, which constraint, and what was received?
Agents retry. Without idempotency, a retry charges the customer twice.
Can a model get from no context to a correct call without asking a human?
Can an agent act for a user, scoped and revocable, with a trace of what it did?
Can you see what agents tried, what failed, and where they gave up?
When it should stop, can it hand back cleanly — or does it just abandon the task?
How it runs
Fixed price on defined scope — never hourly. We're faster than you expect, and hourly billing punishes exactly that.
A few hours with your product. Three specific things an agent can't do, and what it would take to fix each. Yours either way.
Full score with evidence, a prioritised roadmap, and a working proof-of-concept running against your staging.
We build the roadmap. Each phase ships something you own, and you can stop after any one of them.
Start here
We'll spend a few hours driving it the way your customers soon will, and send you what we find. If it's useful, we'll talk. If not, you keep the teardown.
No cost, no pitch, no obligation. Usually back within a week.