Everyone shows you the agents.
This is the machine.
Every AI agent demo you have saved is a screen recording — a glowing dial, a voiceover, one chat window answering one question. So here is the opposite: our own backend, live, counted from the log our agents write while they work. Then score your own setup on the three parts that decide whether it is a system or a show.
Read, not written. These four numbers and the wall below come from
/glass-box.json, refreshed on a schedule from our own run log. No dial on this page is decorative:
there is no fake CPU meter, no fake network graph and no invented event text — because a backend you have to
fake is not a backend.
The wall
The last runs, the mix of work, and the run count for the past fortnight. A run is one agent waking up on a schedule, doing a job and writing down what it did.
Last runs
What the work was
Runs per day last 14 days
Prove it
Every number above, and where it is read from. If a number cannot be computed the tile disappears instead of guessing.
| Number | Where it comes from |
|---|
Job names, client names, file paths and vendor names never leave our workspace: each action is mapped to a plain-English kind and an area of the business before it is published. That mapping is a fixed list, not a judgement call.
A real system is three things
Get these right and it compounds — more context in, more of the business it can carry. Any front end you like can sit on top afterwards.
Several agents, at the same time
One chat window is a person doing the work with extra steps. A workspace means each agent has its own job and its own scope, they run in parallel, and something senior checks their output before it counts.
Your company, written down
What you sell, who buys it, how you sound, how the work gets done — in plain language, in files an agent reads before it touches anything. This is the part nobody wants to do and the only part that compounds.
Allowed to act, on a leash
Real tool access into the inbox, the calendar and the customer list, so it drafts, acts and logs. Behind a human yes on anything outbound or irreversible — and every action written down.
Score your own setup
Answer honestly and you get a number out of 100, the one flag that matters and the single biggest fix. Nothing is sent anywhere. One rule to know before you start: if your agents can act with no human approval and no log, the score is capped at 55 no matter how good the rest is — that is exposure, not readiness.
Workspace
Brain
Access
The scorer on this page is generated from the same code we grade accounts with, and checked line for line: 4000 cases, 36000 comparisons, 0 differences.
Questions
Why do most AI agent demos not prove anything?
Because a demo is a recording. A glowing dashboard, a voiceover and one chat window answering one question tells you nothing about whether anything runs when nobody is watching. The only honest proof is the log: what was done, when, in which part of the business, and who approved it.
What am I looking at on this page?
Our own backend. The numbers and the run list are read from the same log our agents write while they work — jobs that ran, actions taken, what kind of work each one was. It refreshes on its own. Nothing on this page is typed in by hand and nothing is a simulation.
What are the three parts of a system that actually works?
A workspace, so several agents run at the same time instead of you babysitting one chat. A brain, meaning your offer, your customers, your voice and your processes written down where an agent reads them before it acts. And access, meaning it can work inside your inbox, calendar and customer list — behind an approval gate, with every action logged.
Why does the score cap at 55 when I say the agents can act freely?
Because write access with no approval step and no log is not readiness, it is exposure. An agent that can email your list, move a deal or charge a card with nobody checking will eventually do it wrong and you will not be able to prove what happened. Fix that first and the rest of the score counts.
Do I have to buy anything to see my score?
No. The score runs in your browser on this page, nothing is sent anywhere, and you get the single biggest fix in plain English whether or not you ever talk to us.
Can I get a glass box for my own business?
Yes — that is the point of building ours in public. Your version shows what your agents did for you: drafts written, replies handled, leads routed, appointments booked, and the queue of things waiting for your yes. Text Rudy one line and ask for it; a person reads it and answers.
Want a glass box for your own business?
Yours shows what your agents did for you — drafts written, replies handled, leads routed, appointments booked — and the short list of things waiting on your yes. Text Rudy one line and ask for it — no form, and nobody books a call before you know what it is.
Text Rudy about this