Audio brief · 2 min
Practice Guide · Brownfield

RACE Programming on an existing product

Live production. Existing team. No hard cutover required. Two transition paths: choose by your organization's change tolerance.


Why

Why brownfield is harder, and still worth it

Brownfield and greenfield differ by the starting point. Here a lot is already in place, including processes that do not fit the new reality; greenfield starts from nothing. This guide lays out the path from this start and the options along it.

Brownfield adoption faces two constraints greenfield doesn't: an existing team with established habits, and a live product that can't stop delivering. The temptation is to "add AI tools" without changing the delivery model, which produces the 46% AI pilot failure rate (Scrum.org): faster code generation poured into the coordination overhead it was supposed to remove. We measured the effect on one long-running client platform: a developer with agents got 8.3 times faster at coding, and the ten-person team around him delivered 1.65 times as much (The Ceiling of AI Tools).

RACE Programming's value in brownfield is structural change: replace coordination overhead with a spec-first, prototype-gated delivery model. What we see now on the project triangle: ~3× the scope, ~⅓ the cost, or 2-5× faster, worked out from two team shapes. Measured in the agent-to-agent data imports case, against our estimate for a team that already works with AI tools: about a third of the cost of production, 7.1 times less human effort for the same scope, about the same end-to-end speed.

Where the full economics hold: new feature work on the existing product, greenfield features in a brownfield codebase, any work where AI can take a well-formed spec. Where it takes longer: legacy-heavy refactoring with no test coverage, absent or unavailable Team Principal.


What

Two transition paths

Choose based on leadership's change tolerance and available qualified talent.

Either path has one organizer: the TPM, assigned by leadership. The TPM owns the transition plan and its critical path, the order in which teams move, the dependencies between converted and unconverted teams, the risk register and escalation, and the alignment of the sponsor and the Team Principal's stakeholders. The TPM also understands RACE and puts it in place: the right staffing in the right sequence, each person arriving when the work needs them, and a red flag raised early when the project needs something more. The steps below say where the decisions themselves sit.

Fast ROI

Path A: Clean break

  1. Map who can transition. Survey the existing team for AI Fluency or clear intent to develop it. Non-engineering leads (Product Owner, Scrum Master, BA, UX Lead, Test Lead) → AI Product candidates; the Project Manager → TPM candidate; the Product Manager or business customer → Team Principal, client-side. Tech leads, senior and mid engineers → Pit Wall and Pit Crew candidates. Build the roster before forming teams. No qualification = no RACE Programming role.
  2. Form teams, release others. Qualified roster in hand → form one Pit Wall per separate product immediately, with its Pit Crew Quality Engineer from the first story, because a live product is already past its MVP; the Pit Crew Software Engineers follow as the backlog requires. Each product is its own engagement, coordinated by the TPM. Release or reassign those who didn't qualify; that call belongs to line managers and leadership, and the TPM schedules it into the plan. The number of teams follows the number of separate products, not the number of qualified candidates. Start the first story as soon as the first EUS is specified and priced.
  3. Clean break. Faster ROI, cleaner metrics, no cultural bleed from legacy process. Requires leadership support to make the transition non-negotiable. The TPM secures that support from the sponsor before day one and holds it through the cutover.
Lower risk

Path B: Incremental blend

  1. Convert one Scrum team into one Pit Crew. From qualified candidates, in the same order: the Pit Wall, the Quality Engineer, then the Pit Crew Software Engineers as the backlog requires. The other teams continue as Scrum teams on the existing backlog.
  2. Build the machinery, then move work onto the Pit Crew. The Forward Deployed Engineer builds the machinery that lets this one Pit Crew take over the work of the remaining Scrum teams, one stream at a time; a stream moves when the machinery can clear it. The TPM sets the order and puts the dependencies between the Scrum teams and the Pit Crew on the critical path, inside the project budget. People whose stream has moved are reassigned by line managers and leadership.
  3. Full transition when the last Scrum team's work has moved to the Pit Crew. Slower ROI, lower organizational risk, and it works where a hard cutover is politically impossible.

Who

Role mapping for existing team members

Existing role RACE Programming candidate Qualification required
Product Manager / business customer Team Principal Business ownership, budget authority, 1-day feedback SLA
Product Owner / Scrum Master / BA / UX Lead / Test Lead AI Product AI Fluency, EUS authoring, EARS NFR and the Gherkin acceptance tests, owned here; the Pit Crew Quality Engineer automates them. Keeps operational coordination inside the project. Train or confirm on the first stories
Senior / Tech Lead developer Forward Deployed Engineer (Pit Wall) AI Fluency at senior level. Prototype build ≤ 2 days. ADR ownership.
QA Engineer / QA Automation Pit Crew Quality Engineer Owns the executable Gherkin as the test source of truth; sets the guardrails AI agents cannot override
Mid / Senior developer Pit Crew Software Engineer AI-native execution. 4-gate DoD discipline without reminders.
(none, net-new) Silicon Software Engineer The AI agent, present in every Pit Wall and Pit Crew. Not a person to staff: the silicon teammate.
Project Manager (assigned by leadership) TPM Accountable for the project goals: risks, dependencies, critical path, deadlines, phase scope, expectations; understands RACE and puts it in place; organizes the transition. Does not manage people.
No qualification Not assigned a RACE role Reassign or release. Do not form mixed teams.

How

Context mapping before the first story

Brownfield codebases require context work that greenfield doesn't. Before the first story:

  • AI context audit. What can AI read in your codebase without misinterpreting it? Identify: well-tested modules (AI can execute reliably), legacy modules with no tests (AI needs supervision), undocumented business logic (requires human extraction before delegation).
  • Technical debt strategy. Do not attempt to pay down technical debt and adopt RACE Programming simultaneously. RACE Programming's 4-gate DoD prevents new debt. Existing debt is addressed through dedicated EUS items, prioritized by Team Principal as deliberate investment, not a background task.
  • Machinery after the frame. The Forward Deployed Engineer decides when to start building machinery and how much; nobody has to build it before solving the client's problem and setting the frame.
  • Prototype scope. For brownfield features, the EUS prototype may need to integrate with existing APIs or data models. Pit Wall identifies integration constraints in the ADR component of each EUS before prototyping starts.

The first 30 days

Week 1
TPM assigned; transition plan and risk register opened with the sponsor. Team qualification survey. Role assignments. Pit Wall roles confirmed. Team Principal availability SLA in writing, organized by the TPM.
Week 2
First EPR session with Team Principal. Quarterly scope defined. First EUS authored. First prototype built; no production code yet.
Weeks 3 to 4
The first stories and the first releases. Stories flow on Kanban, each pulled when ready. Each story becomes production-ready in two to three days at the floor; releases follow the client's release cadence. Pit Crew executes each approved EUS, the Team Principal accepts each story, and the next Pit Stop releases it to production. Scrum teams (Path B) continue independently on the existing backlog.
Month 2
EUS acceptance rate benchmark. Pit Stop pass rate emerging. Path B: pick the first Scrum team's stream for the Pit Crew to take over; the TPM puts it on the plan.

How to measure success

Brownfield-specific metrics

EUS acceptance rate
% accepted without revision on the first pass, after prototype review
> 80% by month 2
Pit Stop pass rate
% of Pit Stops shipped with UAT green and no rollback or hotfix
> 70% in the first 3 Pit Stops
Rework rate
Stories returned after delivery
< 10% after month 2
New debt introduced
DoD gate failures per story
Zero tolerance after the third Pit Stop
Path B: Scrum streams taken over
Count of Scrum team streams the Pit Crew has taken over
At least 1 per quarter
Honest caveat
Where the full economics hold: greenfield features in the brownfield product, new modules, engaged Team Principal. Where it takes longer: legacy-heavy refactoring with no test coverage, absent stakeholders, contradictory requirements. We'll tell you which side you're on before a proposal.

See the Framework overview for full role and artifact specifications. If your team currently uses Scrum, the From Scrum guide covers role and ceremony translation in depth.

Comments and ideas

If you have tried this route or a similar one, write to me directly on LinkedIn or at paul@raceprogramming.com to share what you have found.

Help putting it in place

If you want help applying this guide in your own team, First Line Software offers that as a service.

FAQ

Frequently asked questions

Can you adopt RACE Programming without stopping production delivery?
Yes. The incremental path converts one Scrum team into one Pit Crew while the other teams continue on the existing Scrum backlog. The Forward Deployed Engineer then builds machinery so that this Pit Crew takes over the other teams' work one stream at a time. Full transition completes when the last Scrum team's work has moved.
How do you assess which team members qualify for RACE Programming roles?
Survey existing team for AI Fluency or clear intent to develop it. Non-engineering leads (Product Owner, Scrum Master, Business Analyst, UX Lead, Test Lead) are AI Product candidates; the Project Manager is the TPM candidate, for project management; the Product Manager or business customer maps to the client-side Team Principal, not AI Product. Tech leads, senior and mid engineers are Pit Wall and Pit Crew candidates. No qualification = no RACE Programming role. Build the roster before forming teams.
What happens to the technical debt in a brownfield project?
RACE Programming's 4-gate Definition of Done (unit ≥80%, integration, E2E, acceptance, all automated) prevents new debt accumulation. Existing technical debt is addressed through dedicated EUS items in the Executable Product Backlog, prioritized by the Team Principal against new feature work, not separately managed.
Who organizes a brownfield transition to RACE Programming?
The TPM (Technical Program Manager), assigned by leadership. The TPM owns the transition plan, the order in which the Pit Crew takes over the Scrum teams' work, the dependencies between the Pit Crew and the Scrum teams still on the existing backlog, the risk register, escalation, and the alignment of the sponsor and stakeholders around the Team Principal. The TPM understands RACE Programming and puts it in place: the right staffing in the right sequence, each person arriving when the work needs them, a red flag raised early. Qualification and reassignment stay with line managers and leadership, raising the budget and priorities with the Team Principal, spend inside the budget with the TPM, the product with AI Product.