Don't transform your company. Build the one that replaces it — starting with one workflow, not a lab or a second company. Map it. Design it again from zero. Cut the design. Build it where the immune system can't reach. Prove it against the old way. Let it earn the work until the edge is the center.
See the SequenceMost methods stitch frameworks together. This one gives each of them exactly one job, and adds the question none of them ask: why your organization will refuse.
You build something with AI, alone, for a day, and hit the wall. Judgment recalibrated before you judge a map. Not part of the ladder — a side door for the executive who wants first-hand evidence before funding anything.
One workflow, not the company. Its value chain on a map, its antibodies drawn on. You see where the waste sits, where the resistance will come from, and how much of your operating model this capability threatens — which decides whether the edge is the right path at all. This map is the baseline: it is never the design.
The mandate, in your words: which capability, built as if the firm were starting today, and what it must do ten times better. One page, signed by you. Whether it drives the marginal cost of supply toward zero is the test the design will have to pass. The loop is named here too: what the workflow learns each time it runs, where that learning leaks today, and which part of it nobody outside the firm could reconstruct. That part is what the edge is built to keep.
The first workflow: highest ratio of coordination to judgment, high-volume, rule-clear, measurable, reversible, low regulatory exposure, and a loop that leaks. No loop, no build. Winnable and reversible beats impressive. And the bar is ten times faster, not ten percent: if the best case is a modest gain, it's optimization, and the method isn't worth running on it. Go or no-go, with a readiness score.
With the people who will run the edge, on a white canvas: if we needed this capability today, with everything now possible, and none of the current organization existed, how would we build it? Then the Cut, on that design — every part, steps, rules, approvals, fields, reports, features, clauses, gets a named owner and a consequence or isn't built. Ten steps or three? The old workflow is never cut; it is retired when the new one has won. Nothing is automated yet.
Only now: your senior people's judgment captured as playbooks, the firm's real trade-offs written down, each remaining task scored for what an agent can take.
Under fifty people, in place. Above, a small twin: three to five of your people plus agents, funded by you, insulated from the unit it will replace. Before handover, red-team the rebuilt workflow three ways — impersonation, injected instruction, exfiltration. Then the new workflow runs in the shadow of the old one, against benchmarks set before the run. Then it earns the work: one percent of it, then ten, then all of it. The parts of the old way that turn out to be needed come back on their own — about ten percent. The old way stays live until there is nothing left to send to it.
Retire the old workflow once the new one carries all the work. Then the next one, easiest first. Re-map quarterly. Every human correction becomes a reason to delete something, not a rule to add. The people move with the work, and the measured capacity created by each workflow funds the next. The edge is now the company.
In a workshop a part is something you can hold. In your company the parts are made of words, which is why nobody counts them. Anything that has a cost is a part. There are two maps on the table: the workflow as it is, which is the baseline and the list of what you will retire, and the workflow designed again from zero, which is the one the Cut runs on. Why is there a wheel? For each part of the new design, two questions, asked out loud: who asked for this, by name? and what breaks if it is gone tomorrow? When the room hesitates, two more: can we live with the risk? and imagine it's gone — do we survive? The smarter the expert defending a part, the harder to ask, and the more it needs asking.
List them for one workflow, not the company. One workflow produces more parts than fit on a whiteboard. The cut design exists before anyone has typed a prompt. And a part nobody owns is a part nobody watches: the same inventory is your attack surface. Why deletion comes first →
The seven categories, the two questions, and the ten-percent check — as a printable checklist you can run on any workflow before your next AI conversation. Free, by email.
Sharing it with someone? Send them samuelpouyt.com/r/parts-inventory.
A way to reinvent a company from one workflow instead of transforming the whole company at once. Map the workflow as it is, design it again from zero, cut every part of the design nobody owns, capture what survives, build it at the edge where the organization's immune system can't reach, prove it against the old way, and let it earn the work until the edge is the center. At company level: don't transform your company, build the one that replaces it. At the level anyone can buy: one real workflow, not a lab or a second company. It combines a value-chain map, the five-step delete-before-you-automate rule, the exponential-company question, and a diagnosis of the twelve antibodies that make transformations fail.
Because a strategy has no override rate, and no loop. One workflow, run in the shadow of the old one against benchmarks set in advance, either wins or it doesn't, and the organization keeps running the whole time. Each move is small, reversible, and self-proving; a failure costs one workflow, not the business.
Step four, and it runs on the new design, not the old process. The workflow is designed again from zero, then every part of that design — steps, rules, approvals, fields, reports, features, clauses, KPIs — is asked two questions: who asked for this, by name, and what breaks without it. Why is there a wheel? Parts with no name and no concrete consequence are not built. The old workflow isn't cut; it is retired when the new one has won, and the roughly ten percent of it that turns out to be needed comes back during the shadow run. Only what survives is captured and automated.
No. Under about fifty people the company is the edge and the method runs in place. Above that, the work moves into a small twin — a few of your people plus agents, funded by the CEO, insulated from the unit it will replace — because the core will reject it otherwise.
Four hours for steps one to three (Awake), ten weeks for steps four to six (Sprint) on the first workflow, then a rhythm of one workflow at a time. The first workflow is chosen to be winnable, not impressive.
When people start asking you to add things back — it shows up while the new way runs beside the old one. About ten percent of the old process returns; if nothing returns the design was timid, if half returns it was a copy. The check that follows is cycle time: the rebuilt workflow should be one to two orders of magnitude faster before anything is automated. If it is ten percent faster, you optimized.
No, and not as a courtesy: it is a rule of the method. The workflow is selected because it has a credible order-of-magnitude cycle-time case, but the shadow run determines the gain. Your people move to the new way with the work, and the first use of any capacity created is more valuable output. Shrinking can always be chosen later and can never be undone.
No. You leave each step with an artifact — a map, a cut design, a working workflow with a falling override rate — not a certificate. Your team is trained by running the new workflow, not by a course about it.
Steps one to three produce the decision before the next programme: the map, the written core-or-edge mandate, and the first workflow with a go or no-go. Awake is CHF 7,500, fixed.
Map What's Blocking YouField notes on stalled AI transformation, executive judgment, and the specification economy — the weekly letter behind Book 2. About once a week. No spam. Unsubscribe anytime.