←→ navigate · F fullscreen
Plan of Attack · Agentic-Coding Enablement

The Mission,Restated.

Two weeks · two engineers · one team · 1.5M lines of 10-year-old legacy.
01Immediate value
Demonstrable wins from agentic coding — inside the two weeks.
02A team that continues
Confident to keep shipping after we leave. The harder half — and the throughline.
Before Anything Else

Define the win first

Work with the PO & EM to name the value and outcomes — before we touch the code.
Cycle timeIdea → merged, before vs. after.
Shipped agenticallyTickets closed with the agent in the loop.
Coverage deltaTests added to the safety net.
Baseline everything in week 1 — no baseline, no before/after.
The Shape of It

The two-week arc

Phase 1 · Days 1–2DiscoveryGoal-setting, persona calls, codebase understanding.→
Phase 2 · Days 3–5We driveHands-on labs & demos; build the CLAUDE.md / skills / MCP stack.→
Phase 3 · Week 2They drive, we coachStandups, office hours, environment tuning, metric tracking.
◆ We drive → they drive — the flip is the whole engagement in miniature.
Phase 1
Days 1–2

Discovery

Days 1–2 · Discovery

Discovery activities

01Kickoff call
Lay out the whole engagement — what we're doing, and why, on day one.
02Calls by persona
EM/PO — 1–2h 1:1s · Engineers — short 1:1s · QA — 30–60min sessions.
03Codebase understanding
Run Claude Code with a discovery skill to map the ground (more on this later).
The persona calls are a prime automation candidate — we'll come back to that.
Days 1–2 · Engineers

Working with engineers

The human pitch: responsibilities shift, they don't disappear. You're driving the race car — when it brakes, you fix it. You don't get out and walk.
Days 1–2 · QA

Working with QA

QA is the sleeper win — converted QAs become the most persuasive advocates on the team.
Days 1–2 · PO & EM

Working with the PO & EM

Negotiate a slowdown
Buy the engineers headspace for week 1 — learning has a cost; name it up front.
Determine success metrics
Existing metrics we can beat — and a gold-mine ticket to point to.
Private read on dynamics
Who's the skeptic, who'll adopt, what's the overall vibe.
Phase 2
Days 3–5

We drive

Days 3–5 · Champions

Work with my champions

Test generation + exploratory planning are the backbone of a durable agentic-coding practice.
Days 3–5 · Skeptics

Turn skeptics into the quality gate

Their skepticism becomes the team's immune system, not a roadblock.
Days 3–5 · Initial Build

Tech setup & initial build

01CLAUDE.md v1
A first hierarchy — repo root down to the gnarly modules.
02Repo skills v1
Skills for the subsystems that need careful handling.
03MCPs
Wire in the tools the team already lives in.
04CI/CD practices
Guardrails for agent-authored changes.
Reviewed with the team at every step — they co-own it, so they can keep it.
Days 3–5 · Sessions

General working sessions

Best-practices call

  • Basic agent use
  • Tools & MCPs
  • Managing the context window

Champion co-led lab

  • Live and hands-on
  • Led with a champion, not just us
  • What to do when things go wrong
Phase 3
Week 2

They drive,
we coach

Week 2 · Coaching

Coaching sessions

Daily standups
Surface blockers; keep momentum visible.
Office hours
Drop-in help, no appointment needed.
Hands-on debugging
Beside them when the agent goes sideways.
Pair programming
They drive the agent; we coach the technique.
Week 2 · Tuning

Tuning the coding stack

Tune the CLAUDE.md's
Tighten conventions where the agent drifted.
Enhance the test suite
Grow the safety net as confidence grows.
Tune the skills
Sharpen what worked; retire what didn't.
Driven by daily-standup feedback — a tight loop, not a one-time setup.
Week 2 · Evidence

Harvesting success metrics

Quantitative · vs. baseline

  • Cycle time — before vs. after
  • Tickets shipped agentically
  • Coverage delta on the safety net

Qualitative · the story

  • An engineer driving solo, on camera
  • Champion testimonials
  • The gold-mine ticket, closed
Feeds the demo and the leave-behind evidence — and we measure work, never people.
Cross-Cutting
Cross-cutting

Key
considerations

Hard Mode, Respected

Codebase size

1.5M
lines of code
10+
years old
many
teams, one repo
Scaling the Playbook

Automation & scalability

01Interview portal / chatbot
Runs the persona 1:1s at scale and gathers the inputs.
02Codebase-discovery skill
Parses structure + git history, finds patterns, maps teams → outputs CLAUDE.md, skills, CI/CD recs.
03Reusable tool stacks
Industry- and company-specific starting points, harvested for the next team.
04Automating pair programming
Feed exported OTel from their Claude Code sessions to an agent armed with our encoded best practices — auto-recommend improvements back to them.
Thank You

Let'stalk.

Questions — and where I'd start on day one.