No True Creativity
It recombines what exists. Your strategy problem is that what exists is what your competitors already did.
This page covers one specific way AI gets things wrong at work, and what to do about it.
It runs in order. What goes wrong, why it happens, where you'd notice it on an ordinary day, who takes the blame, roughly what it costs, and the check that catches it. Then one thing to try this week.
The dollar figures are estimates, not measurements. The assumptions behind each one are printed right there, so you can swap in numbers that fit your job. Anything actually measured carries an OBSERVED tag.
What is no true creativity?
Generation is recombination over learned patterns. This produces genuine value — surfacing options you hadn't considered is useful, and dismissing it is a mistake. But the operation has a structural ceiling: it produces the center of the distribution of documented approaches, weighted by how often they appear.
For strategy, this inverts what you need:
- Frequently documented approaches are frequently attempted approaches. High representation in text means many people did it and wrote about it.
- Competitive advantage requires deviation from consensus. If the move is well-documented, it is well-known, and if it's well-known, it's priced.
- Genuinely novel positions are under-represented by definition. Nobody has written about them yet — that's what makes them novel.
So the model reliably produces competent, defensible, conventional strategy. Which is precisely the strategy that produces average returns, and it will read as insightful to anyone who hasn't seen it before.
This is the mode most likely to be dismissed as philosophical. It is the one with the largest dollar consequence, because it operates on the decisions with the longest time horizons and the least measurement.
What do people assume?
That novel-seeming output is novel — that a strategy the reader hasn't seen before is one the market hasn't seen before.
Output novel to you is drawn from the distribution of what has been written. In strategy, that distribution is the set of moves already made and documented.
Where does it show up at work?
A team uses AI to develop go-to-market options for a new product. It returns five well-reasoned approaches with sound rationale.
All five are standard plays in the category. Three competitors are running two of them. The analysis is competent and the output is a plan to compete on the same axis as everyone else — which is a pricing problem, and pricing problems compound.
Who carries the downside?
Vendor: none. Executive: owns the strategy, eventually. Manager: owns the plan. You: the exposure here is different from every other mode — nobody gets blamed. The plan was reasonable, defensible, and average. This failure produces no incident, no post-mortem, and no learning. It just produces a worse company, slowly.
What does it cost?
[MODELED — not reported]
ASSUMPTIONS Strategic decisions w/ AI input: 12 / year Rate defaulting to consensus: 50% (6) Rate where differentiation was available and forgone: 25% (~1.5) Value of forgone differentiation: $50,000 – $500,000+ (margin premium unrealized, position ceded)
Annualized exposure: ~$75,000 – $750,000
Widest range in Part II, and honestly so. Strategic opportunity cost is genuinely hard to price, and anyone claiming precision here is selling something.
How do you control for it?
Use the model for the base rate, not the answer. Two-step:
1. Ask what the conventional approaches are. Take the output as a map of the consensus — that is what it is genuinely excellent at, and it's valuable. 2. Then ask the human question: what do we know, or what can we do, that makes a non-consensus move available to us?
Step 2 requires knowledge of your specific assets, constraints, and market position — knowledge that lives in your organization and not in the training data. The model cannot do step 2. That is not a limitation to work around; it is the location of your remaining economic value.
CONTROL COST Strategic decisions: 12 / year Structured divergence session: 4 hours each Annual: 48 hours Fully loaded rate: $120 / hour (senior)
Annualized control cost: $5,760
What should you do this week?
RECOMMENDATION
Next strategic question you face, run it deliberately in two passes. Pass one: what does the model say? Write it down and label it "consensus." Pass two: what do we know that the consensus doesn't account for?
Then bring both to the meeting.
You will be the only person in the room who has explicitly separated the conventional answer from the differentiated one — and you will have demonstrated, without saying it, that you understand exactly where the tool's value ends and yours begins.
That is the last mode in this field guide, and it is the argument the whole book has been building toward. AI has made competent conventional work nearly free. The premium moves entirely to judgment about what the conventional answer misses. Part III shows you how to build the controls. Part IV shows you how to prove you ran them.
Evidence
ANALYSIS The entire mechanism section is the author's argument, built from the architecture of generative models plus standard competitive-strategy reasoning. It is not a research finding and must not be labeled as one. This is the most contestable chapter in Part II — and the label is what makes it defensible rather than overreaching.
RESEARCH Model output clustering toward distributional modes; reduced diversity relative to human-generated sets.
RECOMMENDATION The two-pass method is author's practice, offered as advice. Labeled accordingly.
This is one of 24 failure modes. The book gives you all of them — plus the controls that catch each one and a 90-day plan to prove you ran them.
Preorder the Book