AI doesn't widen the response window. It narrows it.

The case for investment is real.

AI compresses the upper part of the stack — Sensing, Acquisition, and Assimilation — by orders of magnitude in some contexts. Threshold-detectable patterns surface faster. External information becomes addressable through retrieval rather than research. Raw inputs convert to candidate interpretations in a working session rather than a working week.

When AI lands well in those stocks, response distance shortens at the front end. Organisations that were Q3 — Instrumentation Gap — sometimes move to Q1 with relatively small investment. Organisations that were Q4 — Slow-Complex Zone — sometimes move to Q2 if the binding constraint was assimilation.

This is the case for AI investment. It's real and worth taking seriously.

The lower stack is governance-bound.

Seizing, Reconfiguring, and Reflexive are governance-bound, structure-bound, and learning-bound respectively. AI does not make a board decide faster than its quarterly cadence allows. AI does not make legacy operating-model coupling release faster than its dependencies permit. AI does not produce double-loop learning where the organisation's culture wasn't asking the question.

When the binding constraint is in the lower part of the stack, AI investment can produce an appearance of velocity at the front end while the back end remains stuck. The signal gets sensed faster; the response still misses the window.

AI is investment-worthy if the binding constraint is in the stocks AI compresses, and a distraction if it isn't.

You can't see it from inside.

Dell'Acqua and colleagues at HBS published the jagged-frontier finding in Organization Science. The study ran on BCG consultants using GPT-4 across structured knowledge-work tasks. Inside the frontier — tasks the model handled well — productivity rose and quality rose. Outside the frontier — tasks the model handled poorly — accuracy fell and, more troubling, consultants trusted the wrong output more than the right one.

The frontier is jagged: tasks that look adjacent are not adjacent for the model. You can't see it from inside. This is the finding that makes AI's contribution to the Augmented Frontier — Q2 — so valuable, and AI's contribution to misclassified problems so dangerous.

Three disciplines that make AI investment land safely.

Each addresses a specific failure mode the frontier produces. None is novel in isolation — what matters is that all three operate together, as a discipline, inside the cadence of the decisions they are supposed to govern.

1. Frontier-position awareness

   Naming, for each decision class, whether you're inside or outside the model's reliable zone. Treating that judgement as a first-class governance artefact, not an unstated assumption. Refreshed at the cadence of the underlying capability shift.

2. Disconfirming-signal authoring

   For each AI-assisted output, producing in advance the specific signal that would indicate the output is wrong. Looking for it. Acting on it when it appears. The discipline lives in writing the signal down before the answer arrives.

3. Anti-confirmation review

   Periodic structured review of where the AI was wrong and the organisation didn't catch it. Honest learning loops, instrumented at the reflexive stock level. The review's value is in admitting what was missed; the format that doesn't permit admission produces no learning.

The window contracts as fast as the distance compresses.

AI doesn't just compress response distance — it also shortens the available window. Competitors deploy faster. Customers shift faster. Markets clear faster. The response distance that was acceptable in a five-year window can be fatal in a six-month one.

A signal that was historically Q1 — Well-Instrumented — can drift to Q3 — Instrumentation Gap — without any change in measured response distance, because the window contracted underneath it. Periodic re-placement under contemporary window assumptions is part of the discipline.

The AI-era SRD diagnostic produces a sharper conclusion: not whether the response is fast enough, but whether it is fast enough for the window the AI-altered market actually presents.