Right now someone describes what they need and hands the search to an assistant. It reads your site fine — the web solved reading twenty years ago. Then it tries to do something: reach a human, submit the form, book the call. That is a different job, and across 40 household-name SaaS sites, hardly anyone has done it. Which is the whole opportunity.
- acme-payroll.comblocked403 to anything that isn't a browser
- betaco.iounreadablethe page needs JavaScript to say anything
- gammahq.coma dead endreadable — but no address and no way to book
- deltapay.comactionablereadable, contactable, booking URL in the HTMLmeeting booked
Three of them were never in the running. The fourth didn’t win on product — it won because it was the only one the agent could finish on.
Accessible to a person is not the same as usable by an agent.
Every row on the left is good design for a human being. Every one of them is a dead end for the software they sent.
None of this is a redesign. The page keeps looking and working exactly as it does for people — you are adding a second set of doors to the same building.
Readable is a solved problem. Usable is wide open.
The same 19 checks we'd run on yours, across 40 household-name SaaS sites. Every number below is measured, not estimated.
The web spent twenty years learning how to be read. It has barely started on being used — so the second one is still cheap to win.
What an agent could finish on those 40 sites
Share of sites where the task succeeded. The short bars are the unclaimed ground. Hover a row for the count.
Your own version of the gap, and the list that closes it.
One scan, three answers, and work you can hand to an engineer on Monday.
The sentence an assistant will say about you
We extract your name, category, pricing, contact and products exactly as a model would, and show you every field it came up empty on. Each blank is a sentence a competitor gets to write instead.
The 7 jobs it tries to finish, and where it stops
Reach a human. Submit your form. Book the meeting. Call your API. Operate your product. Each one is a yes or a no with the evidence underneath, so you know precisely which door is shut.
The work order, ranked by what it buys you
Every gap sorted by points recovered against effort, with a Copy prompt button that writes the ticket for Claude Code or Cursor. Most of the list is a small pull request, not a project.
Four blanks. Each one is a claim an assistant either skips or invents — and a field a competitor filled in.
Or hand the whole list to me.
The faster way to take the lane.
The report is free and the prompts are yours whether or not we ever speak. Most of this list is a weekend of work for someone who already knows the codebase — which is the strange part. The advantage is this cheap only because nobody has taken it yet.
If it is not going to get that weekend, I do it for you. I run the scan, verify every finding by hand, do the work in your repo, and re-scan to prove the delta. You get a before and after score with the diff attached.
If the score does not move, you do not pay.
Agent Readiness Audit. Two weeks, fixed scope, fixed fee — quoted on the call, once I have seen the repo. It is scoped from your code, so I am not going to pretend to price it from a web page.
One person does this work, so there are a few slots a month. If you would rather not hand over repo access, I work as pull requests against a fork. The scan is free forever regardless.
See what an agent can actually finish on your site
Twenty seconds, no signup. You will know within a minute how much of the lane is still open.