Onboarding day 1: how CliAI replaces "ask the senior on Slack"

The first week at a new job is a steady stream of small, embarrassing questions. How do I tail logs in this stack? What's the convention for connecting to the staging DB? Where do I look when a deploy fails? Every senior remembers being on the asking side and remembers fielding those questions five years later. The cost is real on both sides: the new engineer feels slow; the senior loses thirty minutes a day to context-switching.

CliAI doesn't replace mentorship. It does flatten the volume of can-be-self-served questions to nearly zero.

A real first-week scenario

A new engineer joins a backend team. Production runs Kubernetes, staging is a smaller cluster, observability goes through Loki and Prometheus. By Wednesday they need to debug a flaky test that only fails on staging.

Question 1: "Where are the logs?"

$ clai stream the logs from the api deployment in staging, last hour, errors only→ kubectl -n staging logs -l app=api --tail=-1 --since=1h \    | grep -i error

The command shows the structure: where the namespace is, the label selector convention, the time-window flag. Over a week of running similar commands, the new engineer absorbs the team's flavor of kubectl without anyone explaining it.

Question 2: "How do I get into the staging database?"

$ clai port-forward the staging postgres to my local 5432→ kubectl -n staging port-forward svc/postgres 5432:5432

The new engineer learns we don't ssh into pods, we port-forward services. That's a team norm CliAI surfaced naturally.

Question 3: "Why did the last deploy fail?"

$ clai show the events for the api deployment in staging in the last 30 minutes→ kubectl -n staging get events --sort-by=.lastTimestamp \    | grep api | tail -20

Why this is better than a runbook

A runbook is static. The team's stack drifts: someone migrates from Loki to OpenSearch, someone changes the namespace naming convention, someone introduces an Istio sidecar that needs a different exec flag. The runbook stays out of date. CliAI generates against the question, not against a frozen snapshot.

What CliAI can't do is explain the why. It can give the right kubectl command; it can't tell the new engineer that staging is the only environment where the legacy auth path is enabled, and that's why the test is flaky. That conversation still belongs to the senior.

See also

• Refining commands — onboarding is exactly the moment when the first answer often isn't quite right.
• Documentation — the full subcommand reference for when curiosity catches up.