You cannot transform a legacy company from the inside. The work starts when you write that fact as a mandate, not as a line on the homepage.
The Evolut Strategy is simple to recite. Build a governed digital twin at the edge, still inside the firewall, reporting to the chief executive. Rebuild two high-throughput workflows on the intelligence stack. Run them in parallel with the core. Deprecate what loses. Take the next workflow.
None of that survives first contact with the firm as it currently exists.
The core is not stupid. It is doing its job. Efficiency, repeatability, predictability, budget cycles, works-council notice, vendor lock-in, the quarterly close — these are not personality defects. They are the immune system of a company that has learned, correctly, that most change programs destroy margin and leave a slide deck.
If you ask that system for permission to exist, it will give you a steering committee. A steering committee is how twins die.
The immune system is not a metaphor you can coach away
The Evolut Strategy exists because the core cannot be argued into becoming the edge. The pattern does not require villains. It requires an architecture that does not ask the core for permission to exist. Or, paraphrasing boxing philosopher Mike Tyson:
It isn’t plans that separate the winners from the losers, but rather the agility to respond to the unexpected in real time.
In practice, the attack has a short, repeatable repertoire. It looks like governance. It is not.
1. The committee capture. The twin is moved from the CEO’s line to a digital board that meets monthly. Decisions that should take a day wait for the next slot. The parallel run never accumulates enough instances to argue with.
2. The integration demand. CIO and enterprise architecture insist the twin must sit on the official data platform before it may touch a live file. That demand is reasonable as a destination. As a precondition, it is a two-year delay dressed as hygiene.
3. The headcount conversation. Works council, HR, and middle management relocate the work from “which two workflows” to “how many roles disappear.” That conversation feeds the immune system and does nothing on Monday. It also poisons the named owners you need on the gates.
4. The lighthouse swap. The first two workflows are replaced by a visible, politically safe demo — a chatbot on the intranet, a copilot in email, a dashboard the board can screenshot. Throughput collapses. The deprecate-the-loser decision never arrives because nothing comparable was run.
5. The risk appendix. Legal, compliance, and insurance are invited late and handed a model card. They correctly refuse to underwrite an unsupervised junior with production credentials. The project is paused “until the framework is ready.” Guardrails were the product. They were treated as an ethics annex.
6. The budget reversion. The twin is funded from the innovation envelope. In month four, the envelope is cut, and the team is asked to “share learnings with the core.” Sharing learnings is how the edge is absorbed without ever beating the incumbent on evidence.
If you have seen a digital transformation stall, you have seen four of these six. Coaching the people involved does not change the incentives. The mandate does.
What the mandate has to say, in writing
A CEO-owned edge mandate is not a vision statement. It is a short instrument that answers the questions the immune system will ask before it asks them. If it cannot be read in ten minutes and enforced on a Tuesday, it is not a mandate. It is a press release.
Write at least these six clauses.
1. Reporting line. The twin reports to the chief executive, or to a single named deputy with hiring and firing authority. Not to a committee. Not to the CIO as a service request. The C-suite is present at every layer of the stack — only to approve or refuse.
2. Scope that is a chain, not a department. Name the two workflows: input, output, gate, system of record. “AI in operations” is not scope. “Complete RFQ pack to released offer” is. If you cannot name the chain, you do not yet have a twin. You have a slogan.
3. Right to the live file. The twin receives the same inputs the core receives — the live inquiry, the live order, the live PBC pack — and writes drafts into a review queue, not into a sandbox of last year’s examples. A twin that cannot touch production data cannot produce a deprecate-the-loser number. Isolation that looks like safety is how the edge stays a lab.
4. Guardrails as the product, not the appendix. Evaluation suites, traceable logs, rollback, human review queues, named gates. Agents as supervised juniors. Legal and insurance see this stack in week one, not after the first incident. The twin is uninsurable until those controls are what you are buying.
5. The comparison rule. Cycle time, quality, cost, risk — instrumented on both paths, same inputs, for a stated window (four weeks is a serious first pass; ninety days is a serious first verdict). When the twin wins on the agreed score, retire the old path for that workflow. The mandate names who signs the retirement. If that name is a committee, you have already lost.
6. Money and time that cannot be clawed back by the envelope. A ring-fenced budget and a calendar that survives the next forecast. The team is not a loan from the core. Capability transfer is a deliverable, not a phase-out clause that returns everyone to their old desks the week the pilot is “complete.”
What the mandate does not say is as important. It does not promise headcount reduction. It does not promise a platform. It does not ask the works council to bless an architecture. It names two chains, one owner, one comparison, and the conditions under which the loser is deprecated.
Where the immune system is right
A mandate that treats the core as the enemy will fail for a different reason. The CFO is right that an unmeasured twin is a cost center. The CIO is right that an agent with credentials is a non-human identity inside the NIS2 perimeter. Legal is right that residual risk still sits with the legal person. The process owner is right that a first workflow with no named gate is a liability looking for a file.
The Evolut answer is not to overrule those objections. It is to put them in the product.
Humans stay above the loop — yes or no at every material checkpoint, on the record. Judgment is a scarce production function, not a seating plan. The legal person still owns liability, purpose encoded as protocol, the data container, and the learning loop. Decentralized infrastructure, where it appears, is there to protect those assets, not as ideology.
So the mandate should invite the immune system into specific jobs and lock it out of others. Finance instruments the parallel run. Information security and legal design the evaluation band and the rollback.
The process owner is the named gate on the two workflows — which means they can refuse a draft. They cannot refuse the existence of the twin. That distinction is the whole game.
A simple test before you start building
If you cannot put a one-page mandate on the table, do not stand up agents. You will build a chatbot the core can tolerate and call it transformation.
– Can you name the two workflows as input → output → gate → system of record?
– Does a single executive own the twin on Tuesday, without a vote?
– Will the twin see live files in week two, or only after a platform program?
– Are evaluation, logs, rollback, and a human review queue in the first build — not in a later “responsible AI” workstream?
– Is there a written rule for retiring the old path when the numbers say so, and a name who signs it?
– Is the budget ring-fenced past the next forecast meeting?
Five yeses is enough to start. Perfection is how steering committees are born.
What this is not
– It is not a change-management program. You are not asking the core to feel differently about AI.
– It is not a Skunkworks that lives off-site and never touches the P&L. The twin is inside the firewall, on live work, so that when it wins, the retirement is a fact, not a recommendation.
– It is not a promise that nobody will be affected. Coordination costs are falling. The org chart is not the firm. The mandate exists so that what must survive — purpose, liability, the learning loop, human judgment above the gate — is protected while the rest is allowed to lose on evidence.
The first document, not the first model
Seven out of ten chief executives already know there is a high-margin line two people and an agent stack could copy in sixty to ninety days. That line is both an extinction risk and the first place to build the edge.
The 70% question tells you where. The workflow tests tell you which two acts. Neither one will run if the firm’s antibodies treat the twin as a pilot.
Write the mandate first. Then stand up the stack. The model can be fast. The firm stays a firm only if the edge is allowed to exist long enough for the numbers to force a decision — on purpose, on the record, and reporting to someone who can retire the loser.