Audio brief · 2 min

How RACE Scales: Framework and Sketch

Every model for scaling software development grows the same way: it adds people, then it adds teams. Scrum of Scrums, SAFe, LeSS, the Spotify model and plain hiring all work like this. RACE Programming takes a different path. It scales by raising the share of the backlog the machinery clears, and the ceiling is the cognitive focus of each person left in the loop.

Part of this is already in the framework, part is still a sketch. This essay keeps the two apart: first what the framework says, then what is still a sketch.

Part 1: what is already in the framework

The FDE owns the machinery

The Forward Deployed Engineer owns the delivery machinery of their own project. It is written into the role on the framework page. The machinery is the swarm of agents, its integrations with every system the project uses, and the workflows, guardrails and pipelines that clear backlog correctly.

The Pit Wall answers for that machinery, with the FDE as its main engineering member. It keeps the machinery and the project environment working and improving, so they clear more of the backlog.

The machinery has to fit this project. That means the systems in use and the workflow the team agreed on. Above all, it means the workflow that actually works.

Scale comes from the machinery

Every scaling model in software development answers growth with headcount. Scrum of Scrums adds teams and a meeting of their representatives. SAFe groups teams into release trains, LeSS into requirement areas, the Spotify model into tribes. Plain hiring adds people one by one. The size of each step differs from model to model. The move is always the same: more people, more teams.

RACE goes the other way. Adding people to an engagement is a dead end. The path that works is to keep tuning the project environment, the flow of requirements in and results out, and the machinery. Then the engagement takes more of what the client pushes into the backlog.

Applied to RACE, the headcount move would mean more Pit Crews under one Pit Wall, then a split into streams. That is not how RACE grows.

How far the machinery can go on a given engagement depends on model maturity and on how ready the project is.

Once the responsibility sits with the FDE, the question asked of them changes. How much of the backlog does the machinery clear? The rest is cleared by hand: the FDE pulls it onto themselves, or passes it to the people in the Pit Crew. That share is the number to watch.

The headcount path has already been measured. On one long-running client platform, a developer with agents got 8.3 times faster at coding, while the ten-person team around him delivered only 1.65 times as much (The Ceiling of AI Tools). Every person added brings another hand-off, and the speed-up drains away in them. Each hire is paid for at full price, and the client receives a fraction of the gain. What reaches the client is a larger share of the backlog cleared by the machinery.

The RACE Engineer

Building this machinery is heavy. Some FDEs will build it alone. Others will be absorbed by delivery and ask for help. The framework has a role for them: the RACE Engineer, the most senior form of the FDE.

The RACE Engineer sits above projects, part-time and on request, the way DevOps does today. A project that needs one calls one in. Telemetry from engagements is part of the role in every case: what worked in the workflow, what did not, where teams got stuck. The RACE Engineer Owns the Machine explains the role in full.

No universal factory, one north star

The industry is building software factories that take everything in automatically. A universal factory requires rebuilding how the organization works, even for a small project.

What works is a tailored base, introduced into one project and grown there. Today every project has its own machinery. There are as many factories as projects.

The direction is one shared product for the organization’s RACE Engineers. It is a north star, not a current state. It covers more than technology: a standard project environment, disciplined Team Principal, Pit Wall and Pit Crew responsibilities, and the RACE practices.

Telemetry is the instrument for moving toward it. Practices that hold get applied across projects, and projects begin to align. Nobody imposes one factory. How long that takes is unknown. On that path, the organization scales the way each engagement does, through machinery that clears more of the backlog, one project at a time.

Part 2: what is still a sketch

None of what follows is in the framework. Plan nothing on it.

The Quality Engineer moves into the Pit Wall

The execution tier made of machinery is the Silicon Pit Crew. This is a direction, not yet the framework: once the Silicon Pit Crew covers 90% or more of backlog tasks, the Quality Engineer role moves into the Pit Wall.

The reason: the one who writes and the one who checks must not be the same. Separate roles give independent views and a deliberate point of opposition, and a system needs both to produce quality.

With no human Pit Crew left, the check needs an independent home, and the Pit Wall is the one that remains.

Only the Pit Wall remains

In the limit, the structure reduces to the Pit Wall alone. It becomes a micro-team with one specialist per area the project needs: analysis and management, architecture, development, quality assurance.

The limit on that micro-team is each person’s cognitive focus. That is why the systems around it have to call a person in only when a decision is needed.

Both parts point the same way: RACE grows by raising the share of the backlog the machinery clears, one project at a time, up to the focus of the people left in the loop.

Following essays

Three topics get essays of their own. The first is the bottleneck moving to the left and right of the code, into analysis, architecture, acceptance and release. The second is cognitive focus as the limit on each participant. The third is the first practical steps: systems that call a person in only when a decision is needed.

Comments and ideas

If these problems sound familiar from your own work, 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 putting in place what this essay describes, First Line Software offers that as a service.


Written by Pavel Khodalev, author of RACE Programming and CTO of First Line Software. Follow new essays via RSS.