How to Build a Lights-On Agentic Factory

Agents run the line, people keep the decisions

(Available as a one-day workshop or a two-day bootcamp)

Everyone is talking about the dark factory: software built by agents with nobody in the room. Almost nobody will run one. What most teams will build instead, by accident, is a dimmed factory: agents do the work, the lights go down, and a person signs at the end for decisions they never had the time or the evidence to understand. It looks like oversight. It is a signature on a decision nobody understood.

This is a day for building an alternative. A lights-on factory runs unattended wherever a mistake is cheap to put right, and stops, with the evidence laid out, wherever it is not. The wiring is one question asked of every decision an agent makes for you: if this turns out to be wrong, what does it take to put right? The answer sorts decisions into four classes, and the classes tell the factory where to flow and where to stop.

You build it on real work. It is not an introduction: it is taught from the back of the room, with demos in Claude Code and practices that transfer to any tool. Best for groups of 12 to 24, working in pairs.

Want two days? The two-day bootcamp, Building the Twilight Factory, runs this workshop as day one and Agentic Accountability by Design as day two, keeping people close to and accountable for what the factory decides.

Workshop Flow

  • Session 1: Map the Floor

    Map your current agentic workflow

    Mark every step dark, dimmed or lit

    Classify decisions from your own pull requests into four classes

  • Session 2: Lay the Line

    Thin slices and small, falsifiable specifications

    Agents that build to them

    The controls that matter moved out of the model into deterministic checks, gates and hard budgets

  • Session 3: Wire the Lit Stations

    Decide what arrives in front of a person for the decisions that need one

    In what form, and with what evidence

    Add the one agent whose only job is to make sure that person can understand it

  • Session 4: Run It End to End

    Run a feature through the factory, end to end

    Measure tokens, time and rework against the morning’s baseline

    Write a 30-day plan for your team’s factory

For tech leads, architects, platform and engineering leads and senior developers who already use coding agents, and want the gains to hold across a team without burning tokens or people

You’ll leave this workshop with:

  • ✺

    A factory repository routed by decision class, to take home as a template.

  • ✺

    Your own decisions, classified into the four classes.

  • ✺

    A decision-first evidence template for pull requests.

  • ✺

    A 30-day plan for your team’s factory.