When production is live, somebody has to own it.
Building the system is half the problem. Running it is the other half, and most engineering orgs are not staffed to run AI systems on top of everything else they already operate.
Managed Services is what happens after Phase 2. Not a separate consulting motion. The same pod, on a different SLA.
We do not hand you a system and disappear. We do not transition to support through a ticketing queue staffed by people who never saw the code. The engineers who built the workflow are the engineers who keep it running.
Model routing, prompt and policy versioning, eval suites, regression detection, hallucination and drift monitoring, retraining or fine-tuning when the signal justifies it.
Ingestion pipelines, schema drift, source-system breakage, ACL changes, lineage integrity, golden record health.
Tool permissions, action audit, escalation queues, policy updates, replay of failed runs.
Index freshness, recall regression, permission propagation, query latency.
Token spend, model routing, cache hit rate, per-workflow unit economics, monthly cost report against budget.
Access reviews, secret rotation, prompt injection monitoring, model and dependency CVE tracking, audit log integrity.
Connectors to your systems of record, API contract changes upstream, retry and dead-letter behavior.
- Monthly retainer, sized to the surface area we operate.
- Defined SLA per workflow: response time, restoration target, eval threshold, drift threshold.
- On-call rotation that includes named engineers from the pod, not a faceless duty team.
- Monthly operations review: incidents, eval movement, cost trend, adoption metrics, proposed changes.
- 30 days notice in either direction. You can scale up, down, or out entirely. You keep every artifact.
- A written operations report — not a dashboard screenshot — covering eval movement, incidents, cost trend, adoption, and recommended changes.
- A backlog of improvements ranked by business impact, not engineering interest.
- A handover-ready state. If you decide to bring operations in-house, the runbook, eval suite, cost model, and policy files are already in your repository.
- — The workflow is business-critical and a 4-hour outage has a cost.
- — Your internal team cannot reliably staff an on-call rotation for an AI system.
- — Eval drift, model deprecations, or upstream API changes are happening faster than your team can absorb.
- — The cost of a model bill surprise is bigger than the cost of a retainer.
- — You want the people who built it to keep it healthy until you are ready to own it yourself.
- — The workflow is internal, low-stakes, and your team has the cycles to maintain it. We will say so.
- — You want a generic NOC-style support tier. We do not offer that.