For businesses whose AI stopped behaving

Every AI system has a Handholding Ratio.

It's the hours of human attention it takes to get one finished piece of work out of your system. A healthy system runs under 2. If yours is 10, you don't have automation — you have a very expensive employee who can't finish anything without you.

Book a Diagnostic — US$199 / AU$275 Get a free scan first

Nothing to install. No access to grant. Nothing to send. Most Diagnostics deliver next business day.

Your Ratio × your hourly cost × monthly volume = what the broken system costs you, every month.

It's usually a bigger number than our fee. That's the entire commercial argument, and it makes itself.

Does this sound like your system?

You bought AI agents. They negotiate instead of working — every task comes back with questions, or stops at 80% and waits for you to push.
It spends money while idle — scheduled activity, self-checks, self-generated tasks you never asked for.
Someone spends their evenings arguing with it instead of running the business.
"It's basically done" has been the status for three weeks.
Nobody can tell you why it produced that — not without a developer and an afternoon.

If two or more of those are true, the problem isn't the model. It's the architecture around it. That's measurable, and it's fixable. Read the teardown: your folders won't tell you when work stopped →

Your workflow shouldn't be a black box with a dashboard bolted on.

We build AI systems the way well-run infrastructure has always worked: numbered folders are the stages, plain files are the contracts, and "done" is a fact a script can check — not an opinion. Every step exists as a document anyone on your team can open, read, and edit.

invoices/
├── shared/
│   └── invoices.json      ← material arrives
└── stages/
    ├── 01_summarise/
    │   ├── CONTEXT.md     ← the contract:
    │   │                     inputs, process, acceptance
    │   └── output/        ← what it actually produced
    └── 02_dispatch/

You want to…	Typical AI build	GlassBots
Change how a step behaves	Developer, code, redeploy	Edit a text file
See what happened yesterday	Build a dashboard	Open the folder
Prove a task finished	Ask someone	The acceptance check ran green
Explain a decision	Buy an explainability layer	There was never anything hidden
Hand it to a new hire	Weeks of undocumented context	Hand them the folder

How the GlassBots ICM system works — in one picture — The whole idea in one picture.

We don't sell trust in a black box. We remove the box.

You own it. All of it. Including the right to fire us.

Your models, your accounts

We configure your system to run on the AI providers you already pay for. Switching providers is a configuration change, not a rebuild.

Your files, your hardware

The whole system is a folder on infrastructure you control. There's no GlassBots platform to leave, because there's no GlassBots platform.

Your exit, pre-written

Every engagement ships with an offboarding pack: how to revoke our access in one click, and how the system keeps running without us. Published before you ask.

Before anything gets built or fixed, it gets measured.

A fixed-price assessment of your AI system:

A Health Score out of 100 — ten checks, each scored, each with the evidence behind it
Your Handholding Ratio, in hours and in dollars
The three biggest causes, ranked, in one plain-English sentence each
What it costs to leave it alone — monthly and annually
A findings report where every item is marked Verified (we measured it) or Reported (you told us) — nothing presented as measured when it was described

1 · Two minutes, freeRun the scorecard on this site — computed in your browser, nothing sent, nothing stored. The safe first step costs you nothing, including data.

2 · Book at your comfortInterview track: a structured call, NDA first, no files. Folder track: secure upload with a redaction guide — strip customer names; we don't need real data. Files deleted on a stated date.

3 · Decide with numbersSame-day or next-business-day report and a proposal generated from the findings. Worth acting on even if you never hire us.

Health Score: 37/100 — Failing. Handholding Ratio: 9.4 hours per completed item. Top cause: your permission rules make stopping the safe choice — 84% of refused actions were later approved unchanged, so the rules added delay without adding safety.

See the full sample report → — fictional company, real tool, unedited output.

US$199AU$275 inc. GST

Foundation pricing. Includes two free follow-up diagnostic reviews — a founding-client programme with an end date stated when you book.

Book the Diagnostic

Prefer to start free? Take the scorecard — anonymous, nothing stored.

Where we don't fit

We'd rather tell you now:

Real-time, sub-second decisions. Fraud scoring, ad bidding, high-frequency anything. Our architecture optimises for legibility, not latency.
Heavily regulated sectors. Health, financial services, government. These need specialist compliance partners; we'll say so and point you at one where we can.
If nobody on your team can open a text file. You don't need a developer — but you need one person willing to read.
If your system is actually fine. Roughly one in five we assess is healthier than its owner fears. The Diagnostic will tell you that too.

For Australian businesses: from 10 December 2026, amendments to the Privacy Act require businesses subject to it to be able to explain decisions their automated systems make about people. Most systems will answer that with an explainability layer added after the fact. Ours answer it structurally: every decision step is a readable file with its evidence attached — because it was never hidden in the first place. (Scope depends on your circumstances; we'll tell you plainly where it applies.)

Stop babysitting. Start measuring.

One fixed price. One folder. One number that explains where your evenings went — and what they're costing you.

Book a Diagnostic Run the free scorecard

Not ready? Nothing you tap in the scorecard leaves your browser.