"To be everywhere is to be nowhere."
Seneca, Moral Letters to Lucilius, Letter II (translation adapted).
One question will tell you whether your company has a vision. Not whether it has a vision statement; those are universal, like fire extinguishers, and about as often consulted.
When did it last kill something?
Take your time. In my experience the answer arrives slowly, and when it arrives it is usually a decision that was killed by a budget cycle or a lawyer rather than by direction. The statement itself is on a wall somewhere in your building, and it is beautifully made: a sentence about being the leading something, made by people, powered by innovation. It cost real money. It has never once been in the room when a decision was hard.
A vision that has never killed a project is a poster.
Start with what a direction is for, because the wall version hides it. Somewhere in your company this week, a manager you will not speak to is holding a proposal you will never see, and something is about to decide its fate. If the thing that decides is your written direction, you have a vision. If the thing that decides is budget, mood, sponsorship, or a guess about what you would want, you have a statement. The difference shows up as one property: a usable direction lets people at different levels reach the same decision, and explain it the same way, when the author of the direction is not in the room.
Refusal is how you catch that property working. Not because saying no is the purpose of a direction; a direction that only refused would starve the company, and a yes can be direction working too. But approvals are cheap to fake. Enthusiasm, sponsorship, and money can all produce a yes with no direction anywhere near it. A refusal is more revealing, though it is not self-proving, because plenty of things refuse a proposal that are not your direction at all: regulation, cost pressure, a principal's preference, a risk limit, market distress. That is exactly why a refusal has to be tested later against the free-and-legally-clean question and against an authorizing twin. The refusals that count are the promising proposal with a sponsor attached, stopped while it was still attractive, by a company that could have afforded to run it. If you cannot point to one, you may have ambition and a design department.
For a long time this defect was survivable, because something else did the refusing. Execution was expensive. A yes meant hiring, systems, and quarters of waiting, so the sheer cost of doing things killed most of what direction should have killed, and the survivors got called strategy afterward. That governor has moved. Starting has become cheap: a prototype, a small pilot, a card that clears the procurement gate. Finishing has not: integration, security review, retraining, the sponsor's political capital, all of it still expensive and all of it loaded at the end. So the screen that used to sit at the front of the portfolio now sits at the back, and what would once have been refused in a planning meeting is instead abandoned late, after it has consumed attention.
Be careful what you conclude from that, because cheap starts are not the disease. When nobody yet knows which capability will matter, a high start rate is rational, and a company that experiments widely is doing something healthy. The disease is a portfolio with no stopping rule: many beginnings, and no mechanism that says this one does not belong to us, early, while the only sunk cost is small. The cheap era does not ask for fewer starts. It asks for a direction sharp enough to end things on purpose. Your vision was supposed to be that rule. It was on the wall the whole time.
The exercise this chapter runs is different from the ones before it: it runs forward. Somewhere near you right now is a live proposal, attractive, sponsored, undecided. You are going to trace it, from written clause to recorded decision, before anyone knows how it turns out.
Watch a refusal arrive late, in public. In February 2026 WPP announced it would move from a holding-company structure to a single company, organized into four operating units and connected through one platform, after reporting 2025 revenue down 8.1 percent. Its chief executive attributed the underperformance to excessive organizational complexity, the absence of an integrated operating model, and inconsistent strategic execution. Those are the company's facts and the company's explanation. My interpretation, labeled as mine: dissolving your own structure is a real refusal, and the timing is where it gets expensive. What WPP's earlier decision rules were, and why the refusal came during distress instead of ahead of it, the public record does not say. What it does show is the price band for refusing late. This one arrived with a restructuring bill attached.
What changes when the machines start deciding
Every strategy book has told you to have a clear direction, and you have filed that under things everyone says. Michael Porter put the sharpest version in print long ago: the essence of strategy is choosing what not to do. Look at what the current technology adds to that old idea.
Software is starting to act inside companies: placing the order, answering the customer, moving the record. Not all of it, not everywhere, and much of it still drafts and recommends under a human's eye. But wherever a system acts, someone has to write its boundaries down in advance, because a deployed system runs on its instructions, and its instructions are whatever the organization managed to state.
Open those instructions and most of what you find is generic: role, format, permissions, escalation paths, the security and legal scaffolding any competent team produces. Underneath sits a small residue that no engineer, security review, or outside counsel can supply, because it lives nowhere except in your direction. What may this system not do, even when it would work? Whose interest governs when two interests collide? What outcome will this organization not buy at any price? A company that has never stated its refusals finds out, the moment it tries to write that residue, whether it has a direction or a decoration.
There is a public glimpse of the boundary layer, produced under regulation. In May 2025 England's Solicitors Regulation Authority announced that it had authorized Garfield.Law, describing it as the first purely AI-based firm it had authorized to provide regulated legal services in England and Wales, assisting with small-claims debt recovery. The safeguards in the SRA's published account are the interesting part: a client must approve each step, supervision and monitoring are required, named regulated solicitors stay accountable, and the system cannot propose case law. Read that as a picture of what explicit operating boundaries look like when someone insists on them: scope stated, authority bounded, accountability attached to names. Do not read it as evidence that the firm's company vision works. A boundary written to satisfy a regulator demonstrates compliance, which is a different achievement from direction. Boundaries get written for many reasons, and a regulator's demand is one of the sharper ones.
My own record here cuts both ways. The directions that worked for me were short, operational, and visible in the work they stopped; I can still recite them. What I mostly cannot produce is the artifact: a dated record of what got refused, under which clause, at the time. Without one, my recollection is illustration, and recollection flatters the person doing the recollecting. Which is the whole argument for writing the next refusal down while it is still a decision instead of a memory.
One live refusal
The trace goes like this, and you can run it this week on any proposal you are authorized to examine and record.
Choose one live proposal: real, attractive, sponsored, and undecided. The appeal is required, because refusing what nobody wanted tests typing, not direction.
Before the decision happens, write down the exact clause of your direction that governs it. Not the whole statement, the clause: the specific words that bear on this proposal. If you cannot find one, that is a finding. You can stop here; you have learned it.
When the decision is made, record it at the time, with the reason, in writing. Contemporaneous is the discipline. A reason reconstructed after the outcome is known is a press release.
Then ask the hardest question in this chapter: would the answer have been the same if the proposal were free and legally clean? Money and lawyers refuse things all day, competently, with no direction involved. A refusal that survives the free-and-legal question is direction's own work. One that does not was going to happen anyway, whatever the wall says.
Last, name one action the same clause has authorized. A clause that has only ever refused may be a rule, or it may be a justification invented after the fact and applied backward. A clause with an authorizing twin, one real yes and one real no traceable to the same words, has shown it can decide in both directions, which is what deciding means.
The artifact is one line in a ledger: proposal, date, clause, decision, the free-and-legal answer, the authorizing twin. And one check turns the line into evidence about the company instead of about you. Hand the same proposal and the same clause to a manager who was not in the room, and ask what the decision should be and why. If they reach your answer and explain it in the clause's terms, the direction traveled. If they reach a different answer, you have learned that the direction lives in you, which is the most common finding and the most fixable one.
Call this what it is: a bounded probe, one decision deep. It does not grade your company's Vision, and a single traced refusal cannot speak for the other decisions made this quarter. It converts direction from a claim into a record, which is what the rest of this book runs on. A direction you cannot record this way is one you are accepting on faith, and accepting whatever arrives is a strategy that works exactly as long as what arrives is good.
What the refusal changes
One recorded decision changes less than a program promises and more than it sounds like. What it exposes is whether three things that usually never meet actually belong together: the written direction, the decision that was made, and the reason given. When they line up, you have evidence that the direction can operate without its author, which is the property this whole chapter has been circling. When they do not, the gap is specific and fixable: a clause too vague to decide with, a decision made on grounds the direction does not contain, a reason that arrived after the outcome. Each is a different repair, and none of them is visible in a vision statement, however handsome, because statements keep no records. Decisions do.
The person-side consequence follows the same logic, and it is not a score. The professional form of a working direction is judgment you can be held to: decisions made accountably, with the reasoning written down well enough that someone else could learn it, apply it, and eventually inherit the repeatable part. That last clause matters. Recording your reasoning does not diminish you; it makes your old judgment portable, and it frees you to take responsibility for the case the rule cannot settle, which is where judgment grows. A Blank Collar's authority is not the mystery of how they decide. It is the record of what they decided, why, and what they took on when the rule ran out. The record is the authority.
And a dependency was hiding under the whole trace. Every decision you recorded rested on numbers: the proposal's cost, its projected return, the capacity it would eat. A clear no is only as good as the figures behind it, and if the number that justified the decision means one thing to the function that produced it and another to the function that used it, your direction has been deciding on evidence that will not hold still. That is the next condition.