Deployment: one production AI build inside your stack, in 6 to 12 weeks

Deployment is the build stage. Over 6 to 12 weeks I work inside your systems for part of every week and ship the one build the audit recommended: an agent or an AI automation running in production, with its tests, its runbook and a person on your side who owns it.

Key takeaways

  • The scope and the price are fixed at the end of the audit, before any build work is invoiced.
  • You see working software every week, in your own environment, from week one.
  • Evaluation tests ship with the build, so "is it still right?" has an answer after I leave.

What a deployment delivers

  • The build itself, in production: an agent or an AI step inside an existing automation, in the tools your team already uses.
  • Evaluation tests. A set of real cases with known right answers, run before each release and whenever a model changes.
  • A runbook. What the build does, what it may touch, how to pause it, and what to check when it misbehaves.
  • A change log. Every change the agent makes, with the plan and the approver.
  • Handover training for the process owner and for whoever administers the system.

How the weeks run

Week 1
Access through accounts you create, the evaluation cases agreed with the process owner, and a first thin version running end to end.
The middle weeks
One slice of the workflow per week, each ending in a demo. The approval gate and the log are in from the first slice.
The last two weeks
A supervised pilot with real work, fixes from what it shows, then the runbook, the training and the handover.

Length depends on how many systems the build touches and how clean they are, which the audit has already measured.

Pricing a deployment

The audit’s closing report carries the scope and a fixed price for a defined build. Open-ended work can run on a weekly embedded rate instead. I do not publish a range, because the two builds I could quote from differ too much to make one honest.

AI Readiness Audit

$3,500 USD, fixed

2 weeks. Credited in full against a Deployment that starts within 90 days. Invoiced after a short scoping call; pay by card through Paddle or by bank transfer.

Proof from two shipped builds

Auditlab

70 h/month

of audit work saved, with full traceability built into the audit

Audit · plain-language queries over accounting exports, answered inside Jira

Read the case

Leathwaite

5 h/week

saved, and the Excel process it ran on replaced

Professional services · two Forge apps, one feeding Jira automation

Read the case

When to skip the deployment stage

If the audit shows that a configuration fix or an app you can buy closes the gap, there is nothing to deploy and I will not invent something. A deployment is also the wrong step while the process has no owner, or while the data it depends on is still in spreadsheets. Fix those first; a short Systems project usually does it.

Questions

Can a deployment start without the audit?
Only when you arrive with the things the audit produces: a mapped process, an owner, and a checked permission model. Otherwise the first two weeks of the build become an unpriced audit.
Which models and platforms do you build on?
The ones your rules allow. Model providers are chosen to fit your data residency and security requirements, and the build runs in your stack, not in mine.
Who owns the result?
Configuration is yours. For Forge apps and AI engines, handover with source or an ongoing arrangement are both possible, and the choice is written into the agreement before the build starts.

A deployment starts with a scoped recommendation, and that comes from the audit.

About the author

Ben Friedman runs Viter, a forward-deployed AI engineering service. He has spent over ten years running Atlassian and operations tooling, including as internal Jira lead at Aroundtown, and holds the ACP-610, ACP-620 and ACP-120 certifications. He builds JAMES, the AI Jira administrator.