Scenario: Senior Engineer's First 90 Days on a High-Debt, High-Incident Team

Here's my structured 90-day plan.

First 30 days: observe, don't fix

• Shadow on-call for a full week; categorize every page
• 30-min 1:1 with every teammate, manager, product counterpart, adjacent teams
• Read last 20 postmortems, 50 PRs, all ADRs
• Trace one real user request end-to-end through every service
• Produce a written assessment: top risks, quick wins, structural issues
• Frame as questions and proposals, not judgments
• Quick wins that cost nothing: stale doc fixes, obvious logging gaps

Days 30–60: stabilize the bleeding

• Flaky tests: dashboard, quarantine top offenders, root-cause fix weekly, target flaky rate under 5%
• Deploy pain: audit stages, parallelize, cache, cut time 50%, target daily deploys
• Noisy alerts: tune or delete low-signal ones, ensure every alert has a runbook, measure pages/week and MTTR
• Install habits: blameless postmortems, ADRs, runbook-per-alert, design review for new work
• Publish progress weekly — morale and runway for structural work
• Escalate to product if stabilization is blocked — data, proposed investment, shared message with manager

Days 60–90: structural investment

• Debt-reduction track at 15–20% of capacity, named and visible alongside features
• SLOs and error budgets on top 2–3 user journeys — creates shared language with product
• Ownership: every service has CODEOWNER, primary + secondary on-call, escalation path
• North-star metrics: deploy freq, lead time, change failure rate, MTTR, flaky rate, pages/week
• Onboarding doc + buddy system; design reviews; RFC for multi-team work
• Team health: 1:1s again, watch burnout, protect time, celebrate wins, give credit
• 90-day retrospective: what you expected, what you found, what changed, what you'd do differently

Key principles throughout

• Humility first: no one wants technical debt; ask why it got this way
• Strangler pattern, not Big Bang rewrite
• Heroics don't scale — pair and teach, build team capability
• Every postmortem blameless
• Write things down — assessment, proposals, ADRs
• Measure baselines and progress — data anchors trust
• Escalate cleanly, private first, with solution
• Credit outward, absorb small things

Trade-offs

• Stabilize vs rewrite — almost always stabilize; rewrites take 2–3x estimates
• Speed vs process — install 1–2 habits at a time, not 5
• Heroics vs systems — tempting to fix weekend, but team doesn't learn
• Blame vs learning — team was under-resourced, not incompetent
• Escalation vs patience — don't cry wolf; data-backed escalation earns credibility
• Sacred cows — sometimes root cause, handle with allies, small wins, patience

Failure modes to avoid

• The savior — fix everything, burn out
• The judge — criticize before trust
• The firefighter — never improve the system
• The rewriter — lose political capital on Big Bang
• The bureaucrat — too much process too fast
• The silent senior — observe forever, produce nothing
• The hero — solo output, team doesn't grow

Behaviors that work

• Observe first; write it down; measure; pair and teach
• Escalate with data; credit team; admit misjudgments
• Talk to product and management as partners

Red flags to watch

• Manager wants just-fix-it without investment — manage upward
• Product treats engineering as executor — build relationship
• Team has learned helplessness — visible wins + empowerment
• Untouchable sacred cows that are root causes — patience, allies, data
• Broken escalation paths — document, go through your manager

When to leave

• Rare: if org fundamentally refuses to invest in engineering health after 6 months
• Leave cleanly; leave better than you found

Bottom line

• Senior engineer's job is to make the team 20% more effective, not to be the fix-it person
• 90 days is enough to diagnose, stabilize, and lay structural tracks for compounding improvement
• The win is a team that can sustain the improvements without you carrying them