Teaching the double diamond to think
The industry said the design process was obsolete. I kept it, and made it agentic.
A cross-functional lead sits down to frame a client’s business problem. She is a designer, or an engineer, or an AI scientist. The method she needs exists. The frameworks, the decks, the collateral all exist. None of it is anywhere she can reach.
Nobody should rebuild what already exists
Four to five hundred projects a year, and the people on each one working silently in parallel.
The industry’s answer at the time was to retire the double diamond. Board of Innovation’s Stingray model was replacing it. Killing the process was the fashionable move. I kept it and made it agentic — which was the strategic one, because the method was never the problem. Reaching it was.
Duplexity is an operating system for the organisation: the design and behavioural-sciences method, modular AI agents, and a store of frameworks, decks and collateral retrievable at the point of need. Not a tool for designers. A memory for the firm. Built inside a global AI firm as a self-funded practice bet — six months, idea to pilot.
The rules we designed by
One source of truth per client and project.
Every discipline’s vocabulary on the same canvas.
AI does the admin; humans do the judgment.
The human decides when AI speaks.
Four research findings, promoted to rules.
What we did
01 · Research before product. Six practitioners interviewed before a line was built. The finding: they wanted a thinking partner, not an output machine. Spend followed evidence rather than opinion.
02 · Refused the industry’s answer. Stingray was replacing the double diamond. I extended the diamond instead of killing it — one agent per stage, orchestrated across discover, define, develop and deliver.
03 · One home for every client conversation. Every meeting, task and brief in one place, pre-sales through farming. Discovery gets paid for once, then reused by everyone — instead of another point tool in the pile.
04 · Designed the AI’s manners. A second voice, not a first draft. Every suggestion carries its reasoning, and a human checkpoint gates each stage — nothing advances until someone says so.
05 · Shipped it whole. Synthetic users beside real ones, inside the tools teams already trust. Not a demo.
The landscape check found six rivals sitting on research capture — Dovetail, Granola and Elicit among them — and none spanning discover to deliver. The funding figures used in that analysis are public-source estimates from Dealroom, GetLatka, Capterra and company blogs, and are cited as estimates rather than facts.
What changed
Five specialists used to leave one call and write five decks. Now it is one call, one thread, and hypotheses built the same afternoon. Handovers stop killing context. Designers are not replaced; they are multiplied.
5× faster to frame a problem and shape a solution — the vision it was built to, not a measured result. It is shown that way here for the same reason it is shown in grey in the deck.
For the client: no re-explaining, and the stated problem gets reframed to the real one before money moves. For the firm: discovery paid for once and reused.
The honest line
Nothing here was measured. It was a self-funded bet that proved a position and kept the intellectual property — and that, not a performance figure, is what it returned. The thinking now ships as Fluid Intelligence’s Experience Design OS.