Concept · AI Dark Factory Design

You can't automate
what isn't designed.

A dark factory runs without human intervention. Henry designs the step-level specifications that make autonomous operations structurally possible — specifying every step across three planes before removing the human from the loop.

The concept

What is a
dark factory?

A dark factory is an operation that runs without human intervention — the lights are off because nobody is there. The concept originated in manufacturing: fully automated production lines running 24/7 with no human operators on the floor.

Applied to knowledge work and business operations, a dark factory means AI agents executing specified steps autonomously, with human oversight only at defined checkpoints. Customer onboarding that runs itself. Retention monitoring that detects and responds to churn signals without human awareness. Revenue operations that close, invoice, and collect without manual intervention.

The vision is seductive. The problem is that most organizations can't specify what their operations actually do at the step level. They know the outcomes they want. They don't know the atomic steps that produce those outcomes. And you can't automate what you haven't specified.

Henry is the prerequisite. It takes your operation from "we do this generally" to "here is every step, fully specified, with delegation readiness assessed." The dark factory is the destination. Specification is the road.

Manufacturing dark factories

FANUC, Philips, and others run lights-out factories: robotic arms, automated quality inspection, zero human operators. The prerequisite was complete mechanical specification of every step.

Knowledge work dark factories

The same principle applied to business operations. AI agents replace human operators — but only when every step is specified at the level of precision that makes autonomous execution reliable.

The prerequisite

What does step-level specification
look like for autonomous operations?

Every step that will run autonomously must be specified across three planes — with enough precision that an AI agent can execute it without asking questions, without improvising, and without degrading quality.

Work Plane

Intent, inputs, outputs. What this step accomplishes, what it consumes, and what it produces. The strategic "what" and "why" of every atomic unit of work.

Execution Plane

Who performs this step, how it gets done, when it fires, and in what mode — human, AI, or hybrid. The specification that makes delegation possible.

Experience Plane

What the stakeholder should feel. The emotional and perceptual design of each interaction — the layer most operations never specify but always need.

Work Plane for autonomy

Intent, inputs, outputs — defined with machine-readable precision. The autonomous agent doesn't infer intent. It follows the specification. Ambiguity in the Work Plane means failure in the dark factory.

Execution Plane for autonomy

Who performs the step (now: AI agent), how it executes (tooling, APIs, model), when it fires (triggers, schedules, events), and what mode (fully autonomous). Every parameter specified — nothing left to runtime judgment.

Experience Plane for autonomy

What the stakeholder should feel. Autonomous doesn't mean impersonal. The Experience Plane ensures that automated customer interactions maintain the quality bar — tone, timing, and emotional design specified at the step level.

Fit Assessment for autonomy

Ed's Fit Assessment is the gate. A step that passes the assessment with "fully autonomous" delegation readiness is a dark factory candidate. A step that needs human oversight is not — yet. Specification completeness is the path.

The delegation spectrum

Not every step should be
autonomous.

Dark factory design is not about automating everything. It's about knowing which steps can run autonomously and which can't — and why. The Fit Assessment produces a clear recommendation for each step along a delegation spectrum.

Steps at the "fully autonomous" end of the spectrum are dark factory candidates. Steps at the "human-only" end stay manual — for now. The middle of the spectrum is where the design work happens: human-with-AI-assist and AI-with-human-oversight are transitional modes on the path to full autonomy.

The dark factory emerges step by step. You don't flip a switch. You specify, assess, delegate, verify, and then remove the oversight when the specification proves reliable. Henry manages this progression — from human-dependent to fully autonomous — at the step level.

Human-only

Steps requiring judgment, creativity, or relationship context that can't be specified. These stay manual — and that's a design decision, not a failure.

Human + AI assist

Steps where AI handles preparation, drafting, or analysis, but a human makes the final decision. The specification is complete enough for AI contribution, but the quality bar requires human judgment.

AI + human oversight

Steps where AI executes autonomously but a human reviews output at defined checkpoints. The specification is nearly complete — oversight is the last verification before full autonomy.

Fully autonomous

Dark factory steps. Specification complete across all three planes. Fit Assessment passed. Quality bar measurable. No human in the loop. The lights are off.

Where dark factories live

Which AAAERRR stages are
dark factory candidates?

The AAAERRR framework identifies where autonomous operations create the most leverage. High-volume, repetitive stages — particularly Engagement and Retention — are the natural starting points for dark factory design.

The Funnel
Awareness
Acquisition
Activation
The Flywheel
Engagement← HERE
Retention
Revenue
Referral
Off-Ramp
Emergency Exit Sentinel-triggered · Intervention → Graceful Close → Win-Back
Off-boarding Completion-triggered · Engagement → Retention → Revenue → Referral → Awareness

Funnel — Awareness to Activation

Awareness → Acquisition → Activation. A prospect becomes a committed customer. Linear, sequential, non-reversible. If it breaks here, you have a conversion problem.

Flywheel — Engagement through Referral

Engagement → Retention → Revenue → Referral. A committed customer becomes a sustained relationship. Most businesses have a Flywheel problem they're solving with Funnel dollars.

Off-Ramp — two required pathways

Emergency Exit: Triggered automatically by sentinel conditions. Activates Intervention with a time boundary. If recovery fails, Graceful Close executes. Win-Back routes back to Acquisition.

Off-boarding: Triggered at final delivery. A designed linear exit at peak goodwill. Referral advocacy re-enters the Funnel at Awareness and Acquisition.

Related reading

Human + AI Design → AI Isn't Working? → How Henry Works →

Design the specification.
Then turn off the lights.

Start with a diagnosis. End with specifications ready for autonomous execution.

Talk to Henry →