Onboarding after an engineer leaves: keep the context

When a senior engineer leaves, your git history is intact. Your institutional memory is not. It is scattered across Slack threads from 2023, Jira tickets marked Done, PRs nobody reopens, and meeting notes in a shared drive folder named “Misc.” New hires read the code and still ship the wrong abstraction because nobody indexed the why.

What actually disappears when someone leaves?

Departure is not a data deletion event—it is a retrieval failure. The information still exists somewhere; it is just expensive to find and nobody remembers the search path.

• Informal decisions in Slack that never made it to Confluence
• Ticket comments that explain scope changes after the description went stale
• Incident learnings in Meet transcripts not linked from runbooks
• Tribal knowledge about which service to touch first for a class of bugs
• Who to ask—which was always “Alex,” and Alex is gone

Why documentation projects fail after attrition

Teams often react with a “document everything” sprint. It captures a snapshot, not a living system. Two sprints later the docs drift from production and new hires trust them less than asking in a channel that no longer has the expert.

What does Lem AI index for onboarding?

Lem AI (getlem.ai) connects the tools where engineering truth accumulates and makes them queryable with citations—not a single dump into a chat model without sources.

• Slack channels and threads tied to projects and incidents
• Jira and ClickUp tickets with descriptions, comments, and links
• GitHub pull requests, reviews, and commit messages
• Confluence pages and linked Google Drive or Document Hub files
• Google Meet transcripts when meetings are connected

How do new hires use Lem AI on day one?

They ask questions the way they would ask the engineer who left—except the answer comes back with links to evidence.

1. Why did we pick Kafka for event delivery instead of SQS?
2. What broke in billing last quarter and how was it fixed?
3. What did PROJ-220 change in the auth service?
4. Who owns on-call for the payments stack and what is the escalation path?
5. Where is the runbook for partial outage of the search indexer?

A useful answer names the conclusion and points to the Slack decision, Jira epic, or PR that supports it. That is enterprise search with engineering context—not a generic summary that sounds confident and cites nothing.

Why not just use ChatGPT or a wiki search box?

• Public models do not see your private Slack, Jira, or GitHub
• Wikis rarely contain the latest thread where scope changed Thursday night
• Uncited AI summaries train new hires to trust tone over traceability
• Lem AI is scoped to your org’s connected graph with permission boundaries

Onboarding search plus Implementation Agent

Search answers “what happened before.” Implementation Agent answers “what should I build now” when you branch from a ticket—implementation.md carries ticket plus Slack plus docs onto the branch for coding tools.

The same index means a new hire can read how similar features were built, then checkout a branch and get a grounded file for their own ticket. Context compounds instead of resetting every hire cycle.

Practical rollout tips for engineering leaders

1. Connect high-signal channels first—incidents, architecture, team leads—not every social channel.
2. Scope GitHub to production repos and libraries new hires touch in month one.
3. Run a “ask Lem AI” exercise in onboarding week with real tickets from last quarter.
4. Pair search with compliance decision logs so process knowledge is searchable too.
5. Measure time-to-first-meaningful-PR, not just time-to-clone-repo.

Bottom line

Onboarding after someone leaves is not about replacing them—it is about preserving retrievable reasoning. Lem AI indexes where your team already worked and returns answers with sources so the next engineer does not rebuild the same mistakes from scratch.