How to Capture Knowledge Before a Teammate Leaves

The Knowledge Loss Problem

When an engineer leaves, they take more knowledge than is visible at the moment of departure.

The visible knowledge — what's in the code, what's in the documentation, what's in the git history — stays behind. But engineers who've been on a system for two years carry invisible knowledge that doesn't live anywhere else:

Context for decisions: Why the authentication system uses two separate JWT issuers instead of one. Why the database schema has that unusual normalization. Why the deployment pipeline has that specific step that slows everything down but hasn't been removed.

System behavior knowledge: The service that behaves unexpectedly under a specific load pattern. The edge case in the third-party API that only appears in one specific timezone. The database query that's fast in staging but slow in production for reasons that took six months to diagnose.

Relationship and process knowledge: Who to ask about the payment provider's API quirks. Which stakeholder will block the proposal if they're not consulted first. The actual escalation path when the monitoring alert is a false positive.

This knowledge leaves when the engineer leaves. Some of it gets rediscovered — painfully, through incidents and slow onboarding. Some of it is never recovered, and the team operates with invisible gaps for years.

A structured knowledge transfer process conducted before the departure doesn't prevent all loss. But it recovers a significant fraction of what would otherwise be lost, at the moment when the source of the knowledge is still available.

---

The Two-Week Window

Most offboarding periods in engineering are two to four weeks. This is enough time to conduct a structured knowledge transfer if it's treated as a priority from day one of the notice period — and not enough time if it's deferred until the last week.

Week 1: Identify what knowledge exists (the knowledge audit); prioritize what's worth capturing; begin the documentation sessions.

Week 2: Complete the documentation; shadow sessions for high-risk operational knowledge; final review.

This timeline is compressed. It works because the goal is not to document everything — it's to document the knowledge that will cause the most pain if lost and that can't be recovered from other sources.

---

Step 1: The Knowledge Audit

The first step is a structured conversation between the departing engineer and their manager or tech lead. The goal: map what the engineer knows that no one else does.

Knowledge audit questions:

Systems owned:

• What systems or services do you own or have primary expertise in?
• For each: who else on the team can handle incidents and on-call?
• For each: what would a new engineer need to know that's not in the documentation?

Invisible operational knowledge:

• Are there known quirks, gotchas, or failure modes that aren't documented?
• Are there any monitoring alerts that commonly fire as false positives? What's the actual response?
• Is there any system behavior that's different in production vs. staging in ways that aren't obvious?

Context for past decisions:

• Are there significant architectural or design decisions whose rationale isn't documented in ADRs, PR descriptions, or other records?
• Is there any system that looks unusual to a new engineer and would benefit from an explanation?

External relationships:

• Are there external contacts (vendor support, partner engineers, freelancers) that the team will need to maintain?
• Are there any ongoing conversations with external parties that someone needs to continue?

Ongoing work:

• What work is in progress and what context would the next engineer need to pick it up?
• Are there any deadlines or commitments that the team may not be aware of?

The knowledge audit produces a list, not a document. The list is then prioritized by: how much pain will it cause if this knowledge is lost? Can it be recovered from another source (code, documentation, another engineer)? The highest-priority items become the transfer targets.

---

Step 2: Documentation Sessions

For each high-priority knowledge area, a 30-90 minute documentation session between the departing engineer and a teammate.

Format: Screen-sharing walkthrough with live documentation. The departing engineer shares their screen and walks through a system, a process, or a set of decisions. The teammate takes live notes or contributes to a shared document. The departing engineer corrects and adds context as the documentation grows.

This is more effective than asking the departing engineer to "write down everything you know" for two reasons: first, expert knowledge is easier to articulate in response to questions than to enumerate from scratch; second, the session produces a reviewer in real time who can ask "why?" and "what happens when X?"

What to document in each session:

For system knowledge:

SYSTEM KNOWLEDGE DOCUMENT — [System Name]

Overview:
[What this system does; its role in the overall architecture]

Non-obvious architecture:
[Any design choices that look strange without context; 
the reason behind each]

Known failure modes:
[How does this system fail? What are the common incident patterns?
What's the first thing to check when an alert fires?]

Operational notes:
[Quirks, workarounds, things that are true in production but not staging,
versions or dependencies that require specific care]

Historical context:
[Major changes in the past and why they were made;
anything that would confuse a new engineer reading the code or history]

Who else to ask:
[For knowledge about this system that you don't have: who else knows?]

For process knowledge:

PROCESS KNOWLEDGE DOCUMENT — [Process Name]

What this process is:
[What it accomplishes; who uses it]

Step-by-step:
[The actual steps, including the non-obvious ones and common exceptions]

Common failure points:
[What goes wrong; what to do when it does]

Context:
[Why the process works this way; what it was before and why it changed;
who owns it if it needs to change]

---

Step 3: Shadowing for High-Risk Operational Knowledge

For knowledge that's hard to document — operational procedures that require judgment, incident response muscle memory, complex deployment processes — shadowing is more effective than documentation.

Shadow session format: The teammate observes the departing engineer performing the task, while the engineer narrates what they're doing and why. The roles then reverse: the teammate performs the task while the departing engineer observes and provides feedback.

This is high-value for:

• Deployment processes with many steps and potential failure points
• Incident response for a complex system
• Operational procedures with significant consequences for errors
• Any task where the documentation says "use judgment" without explaining what that judgment looks like

One shadowing session per high-risk procedure is usually sufficient to give the team member a working mental model. The documentation written afterward will be more accurate because it reflects what actually happens, not what's supposed to happen.

---

Step 4: The Transition Buddy System

If the departing engineer will be available for some period after their last day (a common courtesy offer), establish a specific "transition buddy" relationship: one engineer on the team who can contact the departing engineer with questions for a defined period (typically 30 days).

The purpose is to handle the knowledge that wasn't captured during the offboarding period — the question that only arises when someone is actually trying to do the thing. This is not an indefinite support obligation; a defined 30-day window is reasonable.

The transition buddy should keep notes on every question asked and answered during this period. These notes become documentation after the 30 days end, when the source is no longer available.

---

Before the Departure: The Ongoing Solution

The best knowledge transfer is continuous, not concentrated in the two weeks before departure. Engineers who follow these practices accumulate transferable knowledge as they work:

ADRs for significant architectural decisions (see How to Write Architecture Decision Records) — so the context for decisions is captured at decision time, not discovered missing at departure time.

Runbooks for operational procedures (see How to Write Runbooks That People Read) — so the "what to do when X happens" knowledge is documented as part of normal operations.

Debugging journal (see How to Keep a Debugging Journal) — so the hard-won debugging knowledge accumulated over months is recorded and transferable.

On-call rotation involving junior engineers — so operational system knowledge isn't concentrated in one person. An engineer who has never seen a specific system fail is harder to transfer to than one who has handled incidents on it.

Teams that maintain these practices consistently find that the two-week knowledge transfer is a completion exercise — capturing the remaining 20-30% that wasn't formalized during day-to-day work — rather than a panic exercise trying to capture 90% of what an engineer knew.

---

Worked Example: A Senior Engineer's Two-Week Offboarding

Setup: Nadia is a senior backend engineer leaving a fintech startup after three years. She's the primary owner of the payments service and has the most context on the card processing integration with a specific payment provider.

Day 1 of notice: Nadia and her tech lead do the knowledge audit. They identify five high-priority knowledge areas: payments service architecture (2 junior engineers know the surface but not the internals), the card processor integration quirks (known only to Nadia), the database partitioning decision rationale (ADR exists but is thin), two ongoing contract negotiations with vendors (Nadia is the primary contact), and the fraud detection system behavior under high load (incident knowledge that isn't in any runbook).

Days 2-4: Three documentation sessions, one per day. Session 1: payments service architecture walkthrough (60 minutes; two engineers participate). Session 2: card processor integration quirks (45 minutes; specifically the timezone edge case and the retry behavior that causes duplicate charges if not handled correctly). Session 3: fraud detection under load (90 minutes; producing a runbook for the specific failure mode).

Day 5: Shadowing session. Nadia deploys to production with a junior engineer observing and narrating; then the junior engineer deploys a test release to staging with Nadia observing and providing feedback.

Days 6-10: Nadia emails her vendor contacts with introductions to her replacement. She updates the three ADRs that are missing rationale. She adds herself to the payments-specific on-call rotation as a secondary for the remaining week so questions come to her directly.

Final day: The knowledge base has grown by 8 new documents, 3 updated ADRs, 1 new runbook, and 1 shadowing session. The tech lead has a list of 4 questions to ask Nadia via email after departure if they come up. Nadia agrees to a 30-day email availability window for the transition buddy.

---

Key Takeaways

1. The visible knowledge stays; the invisible knowledge leaves: decisions context, operational quirks, relationship knowledge, and system behavior edge cases are what's at risk — the knowledge audit surfaces these before the departure.
2. Documentation sessions with a live reviewer are more effective than solo writing: expert knowledge articulates more completely in response to questions than in isolation; a session with narration and live note-taking covers more than an asynchronous "write everything you know" request.
3. Shadowing covers what documentation can't: for operational procedures that require judgment, observing and then performing is more effective than reading; prioritize shadowing for high-risk procedures.
4. A 30-day transition buddy with documented Q&A covers the knowledge discovered after departure: questions that only arise when someone is actually doing the work are predictable; naming a transition buddy and logging the Q&A captures this late-stage knowledge.
5. Continuous documentation practices make the two-week offboarding a completion exercise: ADRs, runbooks, debugging journals, and on-call rotation exposure throughout an engineer's tenure mean the departure knowledge transfer is incremental rather than a crisis.

---

Conclusion

Knowledge loss from engineer departures is predictable and partially preventable. A structured two-week knowledge transfer — knowledge audit, documentation sessions, shadowing for high-risk procedures, vendor relationship handoff, and 30-day transition buddy — recovers a significant fraction of the invisible knowledge that would otherwise leave. The transfers that recover the most are those where the team has been building documentation continuously during the engineer's tenure; the two-week period captures the remainder. The team that treats knowledge transfer as a standard part of offboarding — starting day one, not week two — arrives at the departure date with most of the critical knowledge already secured.

Related reading: AI Knowledge Management in 2025.