Track 08 · Deployment & opsLesson 04 of 11
CI/CD and the agent lifecycle
Ship agents on the same rails as the rest of the estate.
- Specialisation · AI coding & modernisation
- 35 min read
- 75 min hands-on
- Pro
What you'll be able to do
01- You can design a pipeline that builds once, signs, attaches an SBOM and provenance, gates on an offline eval and promotes the same digest.
- You can define a release manifest that pins every behaviour-bearing component, and explain why each is there.
- You can apply semver to an agent release and defend the bump at CAB.
- You can run an agent through lifecycle states, including model deprecation, scheduled re-evaluation and retirement.
- You can fit the pipeline into ServiceNow change records, CAB and standard changes.
Before you start
- Treat prompts like releases: prompts as versioned artefacts.
- Regression tests and CI gates: what the eval gate measures.
- Secrets, dependencies and network isolation: signed images and admission control.
- Registries for models, prompts and agents: the record a release updates.
- Deploying into the customer's cloud and Infrastructure as code for agents: the environments you deploy into.
The short version
An agent's behaviour depends on an image, model IDs, prompt versions, a tool set, a retrieval index and config. If any can change without a release, you cannot say what is running, and your eval describes a system that no longer exists. So you pin them all in a release manifest, hash it, evaluate that hash, and promote it unchanged through the customer's environments and change process. Lifecycle management covers the rest: versions, model deprecations, scheduled re-evaluation, retirement.
- Build once, promote by digest.
- The manifest is the unit of release; eval and promotion records cite its hash.
- Refuse anything that floats:
latesttags, model aliases, unpinned indexes. - Model deprecation is a planned release, not an incident.
- Use the customer's rails. Earn standard changes.
Pro lesson
The rest of this lesson is in Pro.
The mental model, worked example, field note, failure table, exercise and self-check, with the voice tutor.
- Every stage and specialisation, in full
- Every lab and every role project
- Halden weeks 2–10, on your record
- Every drill, live in the playground
- All solutions, plus the server judge
- Lab Assistant and every Teach with AI