AI by Design — A Manifesto in 5 Acts

The problem

Teams default to automating whichever step is easiest to automate, not the one where automation would actually help. A content team adds AI to their writing step when the real bottleneck is brief quality. A product team adds AI to their research synthesis when nobody agreed on what question they were trying to answer. Energy goes into the wrong place, the output improves cosmetically, and three months later nobody can explain why the ROI isn't there.

The deeper problem: AI gets bolted onto workflows that should have been redesigned entirely. The workflow exists for a reason — it's doing a job. When that job isn't named, AI accelerates the process without improving the outcome. You get faster mediocrity.

The smell testIf the project makes sense only because AI can do a particular step faster, the project was always about cost reduction, not value creation. That's a different brief.

The framework

Borrowed from Clayton Christensen's Jobs to Be Done theory and adapted for AI workflow design. The core move: articulate the job the workflow does before touching any of its steps.

Name the job

Complete the sentence: "When we do [workflow], we are trying to [outcome] so that [who benefits in what way]." Not the steps — the purpose. If the team can't agree on the sentence, the workflow has already failed and AI won't fix it.

Audit each step against the job

For every step in the current workflow: does it directly serve the job, does it support a step that does, or is it there by habit? Steps in the third category are candidates for removal, not automation.

Identify AI's role by step

For steps that survive: does AI move the job forward, or does it just speed the step? There's a difference. Speeding a step that doesn't serve the job faster is not improvement. AI earns its place where it moves the outcome — not just the activity.

Protect the human-in-the-loop steps

Some steps are valuable precisely because a human does them. The judgment call, the relationship moment, the creative leap — these are not inefficiencies to automate. Name them explicitly so they don't get swept up in the efficiency drive.

In the workshop

This framework runs as a facilitated session, ideally before any AI tooling decisions are made. The key activity: each team brings a real workflow — something they do regularly, not a hypothetical. The facilitator walks them through the four steps with the actual workflow on a whiteboard or shared canvas.

The moment that always lands hardest is step two. Teams consistently discover three or four steps that exist by habit — no one can explain why they're in the workflow, they've just always been there. The conversation about those steps is more valuable than any AI decision that follows.

A recurring pattern: teams arrive thinking the constraint is "we don't have the right AI tool yet." They leave realizing the constraint is that they haven't agreed on what success looks like. That's a more solvable problem — and a more honest one.

What participants get wrongJumping to step 3 before finishing step 1. The job statement feels too abstract, so teams skip it and go straight to auditing steps. The result is a technically clean workflow audit with no north star. The job statement isn't optional — it's the whole framework.

The artifact

A one-page workflow map with three zones: steps that stay human, steps where AI assists, and steps that get cut entirely. The job statement sits at the top as the filter for every decision below it.

The map is a living document — it should be revisited every time the team considers adding a new AI capability to the workflow. The question is always the same: does this move the job forward, or does it just speed a step?

Ask why the workflow exists before you add AI to it.

Sections

Submit

Colophon