Guide
Last updated: 26 September 2026
The embed is sold on outcomes, not availability. Here is the week-by-week scope Zenovae commits to, and what 'done' means at each checkpoint.
In 30 days a Zenovae embedded AI engineer ships: week 1, access and one visible change; week 2, a production slice with an eval; day 30, a system running in the client's repo with a named owner on their side. Not a slide, and not a vendor-hosted demo.
The engineer gets repo access, joins Slack, and sits in the standups. Before the week ends there is one shipped change the client's team can see — not a discovery deck, not an architecture diagram. The first change is small on purpose: it proves the access, the review flow, and the deployment path all work with a stranger in the loop.
One narrow agent or integration goes live on the client's stack, with an eval for the task it is allowed to do. Narrow is the point: one task, scored, in production. The eval is written before the slice ships, so the first production run already has a success condition attached to it.
Something is running in the client's repo, in their environment, with a named owner on their side. Traces are visible. The failures that block release are written down. This is the checkpoint the engagement is judged on: not lines merged, but a system that operates without Zenovae holding the only copy.
It is not a Palantir-style deployment of a vendor platform — the work happens on the client's stack, and the system stays in their GitHub. It is not a freelancer ticket — the engineer is in the standups, and the three-month minimum exists so the system is not abandoned after the demo. And it is not staff augmentation for non-AI work.
Week 1: repo, Slack, standups, and one shipped change. Week 2: a narrow production slice on the client's stack with an eval. Day 30: a system running in the client's repo with a named owner on their side.
Because access, review flow, and deployment are the real risks of putting an outsider in a repo. Shipping one small change in week 1 surfaces all three while the blast radius is minimal.
An agent without an eval has no success condition, so every later conversation about quality is an argument about vibes. The eval is written before the slice ships so production runs are scored from day one.
The engagement continues on the flat monthly fee: extending the slice, hardening the exception path, and transferring ownership. The system stays in the client's GitHub throughout — there is no vendor copy to reclaim.
The Embedded AI Engineer is a flat monthly engagement with a three-month minimum. The number is stated on the scoping call.
Start a scoping call