A high-margin line that two people and an agent stack could copy in 60–90 days is both an extinction risk and the first edge. Score the line. Then pick the workflow.
Salim Ismail asks every CEO the same question: is there a high-margin line of business that two people with AI agents could replicate in 60 to 90 days? Peter Diamandis reports that seven in ten say yes.
Evolut already treats that line as both a threat and a site: the first place to build the twin. The missing piece is not another telling of the number. It is a way to decide which line, and which two workflows, without a workshop that ends in headcount slides.
The 70% question is a map. Use it as one.
What the question is actually asking.
Not “could AI do some of our work.” That is always true and never useful.
The question is narrower. A line — a profit center with a customer, a price, and a repeatable act. High-margin — enough that a small team would bother. Two people plus agents — not a transformation office. 60 to 90 days — not a three-year program. If the answer is yes, someone outside the industry can stand up a copy without your sales force or your supply chain. They need a stack, a connection, and a quarter.
Our note to founders is the same map from the other side: find the replicable line; build it. The CEO and the founder are looking at one object.
The object is not “the company.” It is a workflow that produces margin and is mostly coordination.
Score the line
Take the candidate lines. Score each from 1 (low) to 5 (high) on five criteria. Write the number in the room. Do not mash them into one vanity total — the shape of the scores is the decision.
1. Margin.
Contribution after the obvious costs. A thin line that looks busy is not the target. A quiet line that funds the rest of the firm is.
2. Coordination intensity versus judgment intensity.
How much of the work is routing, drafting, reconciling, scheduling, filing, checking — the layer agents already compress? How much is a yes or no that should not be reduced to a score? High coordination, low judgment: the line is copyable. High judgment: you may still need a twin, but the human checkpoint is the product, not a courtesy.
3. Data availability.
Can two outsiders reach enough signal from the public web, filings, and customers to stand the line up? Or does the work only run on a proprietary loop you actually own? If the data is already in the market, you do not have a moat. You have a head start you can waste.
4. Regulatory surface.
Licenses, Annex III uses, professional duties, product rules. A high surface does not make the line safe. It changes who can copy it and what the twin must log. Low surface plus high margin is the founder’s favorite cell.
5. Customer trust.
Would the buyer follow a new name if the output arrived faster and cheaper? Or do they buy the legal person, the history, the liability? Trust that is only inertia is not a moat. Trust that is a named relationship and a reconstructable record can be.
A line that scores high on margin and coordination, low on owned data and trust, is the extinction cell. Build the twin there first, or assume someone else will.
A line that scores high on judgment, owned data, regulation, and trust is not “safe.” It is the line you keep inside the legal person while you rebuild the coordination around it.
Two columns
For each scored line, fill both columns on one page. If you cannot fill the right-hand column in an hour, you do not understand the threat.
| Defend this line with a twin | Assume someone else rebuilds it |
| The two workflows we would reconstruct first (named, measurable, high-throughput — not a lighthouse) | The two workflows a two-person shop would reconstruct first |
| What the twin must keep inside the firm: liability, data loop, purpose-as-protocol, human yes/no | What they will rent: model, tools, distribution |
| Parallel-run metrics: cycle time, quality, cost, risk | Their likely offer: faster, cheaper, good-enough, no legacy process |
| Who the twin reports to (the CEO, or is this theatre) | Who their customer already is |
| What we would deprecate if the twin wins | What we lose if they win first |
The left column is the evolut move: a digital twin inside the firewall, two workflows, parallel run, deprecate the loser. The right column is the founder’s reading of the same map, in your handwriting. Both columns have to be honest. One of them will happen.
Workflows, not headcount
Do not start with “how many roles disappear.” That conversation feeds the immune system and does nothing on Monday.
Start with the act that produces the margin. Name it. Instrument it. Rebuild that act on the stack — Sense, Interpret, Decide, Orchestrate, Learn — with a human above the material step. The org chart will follow the evidence, or it will fight the evidence. Either way, you will know.
Two workflows, not twelve. High-throughput, not politically safe. If the first two cannot beat the old path on the four numbers, you do not have an edge strategy. You have a pilot.
A simple test
If two people published a price for this line on Monday, which two workflows would they ship by the end of the quarter — and do we already own a twin of those two?
If you cannot name the workflows, the 70% answer was a mood. Score the line. Fill both columns. Then build, or admit you are waiting for someone else to read the same map.