The engagement
Eight weeks to an operation that runs
Gibreen is delivered, not downloaded. A fixed-scope engagement that starts with a measurement and ends with one workflow genuinely running — and a second one under way — inside your own instance.
Why it is delivered rather than sold
The most-cited study of enterprise AI adoption found that the small minority of projects producing measurable return had one thing in common. They were not the ones with the best models or the biggest budgets. They were narrow, embedded in a specific workflow, and bought from a specialist who implemented them.
The failures were not model failures either. The named causes were cost that ran away, value nobody could point to, and risk controls that were not adequate. All three are implementation failures. So the implementation is the product, and the software is how we deliver it repeatedly.
You can stop at the end of any week and keep what was built. The charters, the record, the connector configuration and the workflow definitions are files in your repository. There is no week in this plan where walking away leaves you with nothing.
The eight weeks
-
Week 1
Baseline and boundaries
We measure the thing you want fixed before touching it: how long it takes now, how often it is wrong now, what it costs now. Without this number, nobody can honestly tell you in week eight whether it worked.
In the same week: the data processing agreement, the list of systems in scope, and the list of acts that will never be automated. The refusals are agreed first, in writing, while nobody is under delivery pressure.
-
Week 2
Your instance, and read-only access
An isolated instance is provisioned for you. Connectors are configured read-only against your systems. Nothing can write yet, by construction rather than by policy — which means week two is the week you can afford to be relaxed about scope.
-
Week 3
The first role, written down
We write the charter for the first agent with you, in sentences: what it is responsible for, what it may reach, what it must ask about. If you cannot explain the role to a new hire in one page, it is not ready to be an agent either — and that discovery is worth the week on its own.
-
Week 4
It runs, and you watch
The agent runs on a cadence, reading and drafting only. Every output is reviewed. This is the week that produces the honest hit rate, and it is the week where roughly a third of scoped workflows get redefined — because watching the work is how you discover what the work actually is.
-
Week 5
The first gate opens
One act — a specific one, the smallest useful one — becomes possible, behind the gate. The agent proposes, you approve, the act goes through your application's own logic. The first approved write is a milestone with a date on it, and you are standing there when it happens.
-
Week 6
Volume, failure, and recovery
We deliberately run the failure cases: the system is down, the data is malformed, the model is rate-limited, the answer is ambiguous. What a system does on its worst day is the only part of it worth testing, and this is when we find out what your agent does at 2am when nobody answers a gate.
-
Week 7
The second role
Built by your people with us watching, rather than by us with your people watching. If the second one is harder for you than the first one was for us, the engagement has failed at its actual purpose.
-
Week 8
The measurement, and the decision
The week-one baseline, re-measured. What it costs to run per month, from the real usage ledger, not an estimate. What the gates caught. Then the honest conversation: continue, narrow, or stop — with the numbers on the table rather than a renewal conversation.
What you get, concretely
A running instance
Isolated, yours, with your agents in it and their charters as files you can edit. Hosted by us, or on your own infrastructure if your procurement requires it.
Two workflows live
One built by us, one built with you. Both gated, both measured against the week-one baseline.
The record from day one
Every run, decision and refusal since week two, in a repository you hold. This is the artefact an auditor will ask for, and it exists whether or not they ever do.
The numbers
Cost per run and per month from the real usage ledger. Gate latency — how long your people actually take to answer. The hit rate before and after.
What we need from you
Three things, and the engagement does not work without them. We would rather disqualify a project in week zero than discover this in week five.
A named person at the gate
Someone with the authority to approve the act and the availability to do it within a day. Not a committee. This is the single strongest predictor of whether the project works.
Access, in week one
Read credentials to the systems in scope, and a sandbox or test instance if one exists. Access that arrives in week four costs three weeks of the eight.
One workflow, chosen and defended
Narrow beats interesting. The workflow that is boring, repetitive and currently annoying somebody every week will outperform the impressive one every time.
If nobody will answer a question within a day, the gate becomes a queue, the queue becomes a backlog, and the agent becomes something everybody works around. We will say this on the first call, and we will say it again in week one.
After the eight weeks
The instance keeps running on a per-instance subscription. What that covers, and what it deliberately does not, is on the pricing page.
Additional roles and workflows are scoped separately, and by then you will have built one yourself and will know roughly what it should cost. That is intentional: a client who can estimate the work is a client who cannot be quietly overcharged for it.
Talk to us
Start with a conversation, not a signup
Gibreen is not sold from a pricing table. Tell us which system of record you run and which job is eating your week, and we will tell you honestly whether this is a fit — including when it is not.