A market monitor can run every Tuesday for five months and change nothing. The feeds arrive. The model summarizes them. Nobody can name the claim that moved a price, a watchlist, or a meeting.
SENSE exists for that diagnosis.
SENSE is Len P. van der Hof’s five-phase design grammar for strategic agent systems: Scan, Evaluate, Navigate, Simulate, Execute. Give it one strategic question and it makes the team specify how evidence enters, becomes a claim, changes course, meets a rehearsed future, and arrives where a person can decide.
It comes from AI Agents for Startup Strategy, which is available now, and it is the book’s own synthesis. The book credits its ancestry, OODA’s tempo logic and Weick’s sensemaking among it, and then departs from both: SENSE is prospective and prescriptive, a builder’s checklist rather than a theory of cognition. It is not an industry standard and not a validated assessment scale.
The layer and the grammar
A strategic intelligence layer is the standing system: named inputs, sourced claims, a decision cadence, owners, and expiry rules. SENSE is one grammar for designing and diagnosing that system.
The distinction keeps the architecture honest. A model, a feed, or a dashboard is a component. The layer exists when information has a governed route into judgment. SENSE lets you inspect that route one phase at a time.
The five SENSE phases
| Phase | Required output | Collapse mode |
|---|---|---|
| Scan | A bounded set of signals tied to one strategic thesis | Collecting whatever is easy and producing a feed nobody reads |
| Evaluate | A sourced claim with confidence and contradictions visible | Summarizing findings while exporting all judgment back to the reader |
| Navigate | A change in attention, routing, assumptions, or preparation | Watching in month six exactly what mattered in week one |
| Simulate | A small set of plausible futures with assumptions and signposts | One confident forecast wearing the word scenario |
| Execute | Work delivered to a named owner, deadline, and decision moment | A polished digest sent into no meeting and no action path |
Scan is collection with a thesis. Start from the exposure: what would have to change in the world to damage or invalidate the strategy? That question decides which competitor pages, regulatory sources, customer signals, or internal measures belong in scope, and it excludes the rest.
Evaluate turns signal into a claim. The output says what the evidence means for the named exposure. It carries sources, a confidence mark, and any contradiction with the current position, stated rather than overwritten. A summary is raw material. It has not completed Evaluate.
Navigate changes what happens next. Tighten the watchlist, raise the review cadence, route a question to the founder before the meeting instead of after it, revise an assumption. Navigate is the steering function. It makes the layer a loop instead of a pipe.
Simulate rehearses futures. Test a small set of plausible states, the assumptions that produce them, and the signposts that would tell them apart. The phase prepares conditional action. It does not predict one future with borrowed precision.
Execute meets a decision or it never happened. Deliver the brief, the alert, the draft response, or the recommendation to the person and the moment that can use it. In strategic work, Execute is the smallest phase in autonomy terms. It usually means prepare and present, not act unilaterally.
Why SENSE has no Decide phase
The omission is deliberate. SENSE prepares a choice and routes it into a decision cadence. A person keeps judgment, under a written delegation boundary.
That boundary sorts decisions into lanes. Some low-cost, reversible actions may run on their own. For consequential calls, the system assembles the claim, the evidence, and the options, and a person makes the call. Some decisions stay reserved for human framing from the start. None of those lanes turns Decide into a hidden sixth box.
The boundary matters most when the output is excellent. Fluency and a complete evidence packet make rubber-stamping feel like review. Naming who decides, before the system has earned trust, is what keeps review real.
Partial systems can be honest
Not every useful system needs machinery in all five phases. A narrow regulatory monitor may Scan and Evaluate, then hand Navigate and Execute to a weekly review that already exists. A scenario system may start from human-curated evidence. Those are sound designs, and often the best available ones.
The test is whether the gap is intentional and the handoff has an owner. “Simulate is deferred to the quarterly session” is a design choice. A project team assuming that someone else runs Simulate is a design failure.
This is where SENSE becomes a diagnostic rather than a slogan. “The AI project disappointed us” is a mood. “We built Scan five times and never specified Evaluate” is a repair order.
Run the five-line diagnostic
Write one strategic question at the top of a page. Then answer:
- Scan: Which sources matter, and which thesis puts them in scope?
- Evaluate: What claim shape, evidence standard, and confidence mark will the system produce?
- Navigate: What may change when the claim changes?
- Simulate: Which futures and signposts must stay ready?
- Execute: Which owner and decision moment receive the work?
Mark any phase handled by a person, another system, or a later cadence. Blank is allowed when it is deliberate. Invisible is not.
What is an evidence ledger? shows one place an evaluated claim can live. SENSE: a 90-day strategic intelligence layer covers installation, gates, and artifacts. This page stops at the definition: five named phases, one human decision boundary, and no unread summary pretending to be a layer.
Cite this:What is SENSE? Five moves, not another unread summary.Len P. van der Hof. https://lenvanderhof.com/en/blog/what-is-sense/ ·