NUMBERS FIRST
AI agent development cost comes down to four things: how many processes, how many systems they touch, how many exceptions they carry, and whether you want us to keep running it. This page explains all four, and the shapes we sell them in.
What drives AI agent development cost
AI agent development cost is set by scope, not by seats. The four drivers are the number of processes in scope, the number of systems each one touches, how many exceptions the process carries, and whether you operate the result yourself or we do.
Two of those surprise people. Integration count matters more than agent count, because every system brings its own authentication, rate limits and failure modes. And exception density matters more than volume, since a high-volume process with three exception types is cheaper to automate than a low-volume one with thirty.
The readiness audit prices your build, and it does it before you commit to one. That sequence is deliberate.
How an engagement is shaped
Three ways to start. Every one of them begins with a readiness audit, and the audit is what produces your numbers.
Start here
Readiness Audit
Fixed scope, 3-10 days
We map your workflows and score every candidate process on value, feasibility and risk.
- Workflow mapping across the departments you name
- Every candidate process scored on value, feasibility and risk
- A written blueprint for the first agent
- Cost and time-to-value estimate for that agent
Most common
Build
Scoped from the audit
One process, built to production: integrations, orchestration, retries, observability and deployment.
- Everything in the Readiness Audit
- Prototype on your real data, measured for accuracy, latency and cost per task
- Production build with integrations and orchestration
- Eval suites, approval gates, audit logging and kill switches
- Shadow, assisted and autonomous release stages
Run it for us
Managed Hive
Retainer, billed monthly
We operate the hive: watch the evals, swap in cheaper models that still clear the bar, repair drift and report every month.
- Everything in Build
- Continuous evaluation and drift repair
- Model routing reviewed against cost and quality
- Monthly report on volume, pass rate and cost per task
- New agents added as processes qualify
| Included | Readiness Audit | Build | Managed Hive |
|---|---|---|---|
| Readiness audit | Included | Included | Included |
| Blueprint and autonomy boundaries | Included | Included | Included |
| Prototype on your data | Not included | Included | Included |
| Production build and integrations | Not included | Included | Included |
| Eval suites and approval gates | Not included | Included | Included |
| Ongoing operation and drift repair | Not included | Not included | Included |
| Monthly reporting | Not included | Not included | Included |
Figures are agreed in writing before any build starts.
THE MODEL
Per task, not per seat
Seat pricing punishes you for the thing you bought. If a hive works, more of the process runs through it, and a per-seat model would charge you more for using it less by hand.
So the Hive retainer is set against cost per task: the model calls, the compute and the share of operation attributable to that work. It falls when we route a task to a cheaper model that clears the eval bar, and that fall is visible in your monthly report rather than kept by us.
Scout and Forge are fixed. You know the number before the work starts, and a change in scope is a conversation rather than an invoice you find out about later.
The comparison that matters is not our fee against another vendor’s fee. It is cost per task against the loaded hourly cost of the person doing that task today, and you can model that yourself.
Which shape fits
| Measure | Scout | Forge | HiveRecommended |
|---|---|---|---|
| You are asking | Should we? | Build it. | Run it. |
| Commitment | Fixed fee, days | Fixed scope, weeks | Monthly, ongoing |
| Ends in | A decision | Agents in production | It does not end |
| Who operates it | Nobody yet | Your team | Us |
| Evals and monitoring | Designed | Built and handed over | Watched and repaired |
| New agents added | Recommended only | In scope for one process area | Each quarter |
| Exit | It is already over | You own everything built | Notice period; you keep everything |
| Most clients | Start here | Go here next | Decide after release |
BEFORE YOU ASK
What we do not cover
Your model and infrastructure spend
Model calls, compute and storage are billed by your providers to you, at cost, visible in your own account. We do not resell inference at a margin.
Third-party licenses
If a process needs a tool you do not already hold, that is your line item.
Your team’s time
A process owner, a reviewer during release and someone who can approve credentials. It is a few hours a week, and the build slips without it.
Change management
We will tell you what changes about a job. Telling the person whose job it is remains yours to do, and doing it badly is the most common reason a working agent gets switched off.
Data remediation beyond the agreed scope
Sometimes the audit finds that the data has to be fixed before anything can read it. We will say so, and quote it separately rather than absorb it quietly.
Put your own numbers in
Volume, minutes per task and how much of it is routine. The model shows hours returned and the assumptions behind them, including the ones it deliberately does not count.
Frequently asked questions
Why is there no price on this page?
Because we would have to invent it. AI automation pricing depends on the process, and the honest range across the work we do is wide enough that publishing it would tell you nothing. The audit is deliberately cheap and fixed-fee for exactly this reason: it is the shortest path from "what would this cost" to a real number for your process, and you can stop there.
Can we get a rough figure before the audit?
Yes, on a call. Describe the process, the systems it touches and roughly how often it runs, and we will give you a shape within the conversation. It will be a range, it will be labeled as an estimate, and the audit is what narrows it.
What happens if the build costs more than expected?
Forge is fixed-scope, so the risk of the build sits with us rather than with you. What changes the number is a change in scope, and that is a decision you make with the revised figure in front of you. Nothing gets absorbed silently and then reappears as a variation at the end.
Do we have to take the retainer?
No. Plenty of clients build with us and operate it themselves, and everything is written to be handed over: the runbooks, the eval suites, the dashboards. The retainer exists because somebody has to watch the evals and repair drift, and most teams would rather that was not them in month four.
How does the cost compare to hiring?
That is the right comparison, and it is the one the calculator makes. A hire is a fixed annual cost with fixed working hours. A hive is a build cost plus a running cost per task, at any hour. Which wins depends on your volume and your loaded hourly cost, which is why we would rather you modeled it than took our word for it. See what reference builds looked like.
Is there a minimum commitment?
Scout has none beyond itself. Forge is a single fixed-scope project. Hive runs monthly with a notice period set in the agreement. We do not sell multi-year lock-ins, because a hive that is worth keeping does not need one.