← Back to Resources
Who this role is for: A senior engineer or staff-level technologist who enjoys teaching, writes cleanly, and reads both domain and code fluently.
Who it is not for: A librarian, a gatekeeper, or a process owner. The Context Manager is a coach with a living set of context files, not a custodian of documents.
Pairs with: The Delivery Model,
Per-Feature Context Folders.
What the Context Manager owns
The Context Manager owns the health of the shared context. The shared context is not a document store — it is the set of machine-readable files every agent session reads before touching your codebase. If it drifts, every squad pays the tax in subtle rework. If it is crisp, every squad compounds.
The Context Manager does not write every file. They make sure the right files exist, that the files say what the team actually means, and that when confusion surfaces it flows back into the context files rather than into the next sprint.
The four parts of the job
None of these is a project with an end date; all four run continuously. A good week touches all of them.
Part 1
Maintain
Keep the shared context accurate, week in, week out.
Weekly audit
- Run the audit: stale files, contradictions, broken cross-references, files no squad touched since last review.
- An agent scans; the Context Manager triages.
- Each finding is either fixed, filed as a coaching topic, or escalated to the owning role.
What good looks like
- Every file has an owner, a size target, and a review cadence.
- Contradictions between files are caught within a week of appearing.
- No squad reports “I couldn’t find the answer in the shared context” without that gap becoming a new entry or edit.
Part 2
Unblock
Event-driven: someone hit a wall.
Triggers
- A squad asks a question the shared context should have answered.
- An agent suggested a wrong pattern and the Squad Lead flagged it.
- A reviewer caught a recurring class of bug.
How to respond
- Find the root cause: a content gap, a structural issue, or a skill gap.
- If content: edit the file that should have had the answer, and add a forward reference from adjacent files.
- If structural: propose a change to the formatting standards or the file layout.
- If skill: pair with the engineer on reading the shared context well.
Part 3
Improve
Capture what works before it fades.
What this looks like
- A new pattern emerges in one squad; the Context Manager writes it up as a candidate standard.
- A recurring workflow becomes a template in the formatting standards.
- Short clinics show squads how to adopt the new pattern.
What good looks like
- New patterns land in the shared context within a week of being proven in a squad.
- The shared context grows by consolidation, not sprawl — new patterns replace old ones when they’re better.
- The audit log shows a steady cadence of improvements, not long quiet stretches.
Part 4
Coach
Embedded in every interaction.
Where coaching happens
- Every fix is also a teaching moment: the Context Manager explains the principle, not just the change.
- Pair sessions with Squad Engineers on how to brief agents and read context files.
- Reviews of agent PRs where the Context Manager calls out context-vs-execution mistakes.
- Office hours and short clinics on context-first thinking.
What good looks like
- Over time, the Context Manager is asked fewer of the same questions — the team internalizes what the shared context should say.
- Squad Leads take on progressively more responsibility for context health in their own area.
A sample week
One way to structure the week so all four parts of the job get attention. This is a starting point — tune it to your program’s cadence.
| Day |
Focus |
Parts in play |
| Monday |
Weekly audit. Run it with an agent, triage findings, file follow-ups. |
Maintain |
| Tuesday |
Unblocking day. Address open context-gap escalations from squads. Fix the files, not the symptom. |
Unblock, Coach |
| Wednesday |
Pair sessions. Sit with one or two Squad Engineers as they brief agents and review output. Catch context bugs in flight. |
Coach, Unblock |
| Thursday |
Improvement. Write up the newest pattern that emerged. Propose the template. Circulate for feedback. |
Improve |
| Friday |
Context retrospectives with each squad. What confused an agent this week? What context change would have prevented it? Pair with the Squad Lead on the top one or two items. |
Maintain, Coach, Improve |
Making the work visible
Much of the Context Manager’s work is background work — the moments of confusion that didn’t happen. Metrics make it visible so the role doesn’t get questioned or cut.
Context audit score
Proportion of files that passed the latest audit (no stale, no contradictions, no broken refs). Trend is what matters.
Time to answer
Median time for a squad to answer a common domain question by reading the shared context. Falls as the context matures.
Context gap escalations
Per-month count of escalations from squads that a context edit resolves. A rising number means the work is visible; falling means it’s effective.
Onboarding time
Days for a new engineer to land their first agent-assisted PR. Falls as the shared context covers more.
Pattern adoption
Proportion of squads that have adopted the last three patterns published by the Context Manager.
How to know the role is working
Agents get less wrong over time. The same class of mistake stops recurring because the shared context explains it explicitly.
Onboarding accelerates. New engineers can answer their own questions by reading the shared context, and their first AI-assisted PR lands quickly.
Squad Leads coach their own squads on context. The Context Manager becomes a teacher-of-teachers rather than a single point of knowledge.
Warning signs. The same question being asked three times. Agents proposing the same wrong pattern across squads. Nobody mentioning the audit log. A squad that never surfaces context gaps — that usually means they’ve stopped reading the shared context.
← Back to Resources