I Gave Claude's Newest Model a Strategy Problem. It Wrote Beautifully. It Was Still Wrong.
Share
I Gave Claude's Newest Model a Strategy Problem. It Wrote Beautifully. It Was Still Wrong.
Anthropic launched Claude Fable 5 in June — a frontier model that tops finance and analytical benchmarks, holds coherent reasoning across a million tokens, and reportedly compressed a two-month engineering migration into a single day at Stripe. It is, by every technical measure Anthropic publishes, the most capable general-purpose model they have ever released.
None of that makes it good at building your strategy. And the reason why is more interesting than "the AI isn't smart enough yet."
The comparison everyone gets wrong
When a new frontier model launches, the instinct is to ask whether it can now do the job a specialized method used to do. Can Fable 5 replace a strategy consultant? Can it replace a framework like the one I built into TheStrategist.me?
That question has a hidden assumption baked into it: that strategy work is a capability problem. That the reason AI-generated strategy documents used to feel shallow was that the model wasn't smart enough to go deep. Give it more reasoning, more context, more benchmark-topping intelligence, and eventually the gap closes.
I tested that assumption directly. I ran the same strategic question through unguided Fable 5 and through TheStrategist.me's sequenced method — same business, same starting information — and graded both against ten dimensions that determine whether a strategy actually holds together: problem framing, analytical rigor, customer specificity, competitive defensibility, strategic choice, execution clarity, measurability, organizational realism, communication quality, and sustainability of the review loop.
Unguided Fable 5 scored 54 out of 100. TheStrategist.me scored 100.
That gap is not a capability gap. It is a sequencing gap — and understanding why will change how you think about every AI-assisted strategy session you run from here on.
What a smarter model actually gets you
To be fair to Fable 5, the capability jump is real, and it matters. A more capable model writes a better single SWOT than a weaker one. It extracts numbers from a messy PDF more reliably. It holds a thread across a longer conversation without quietly contradicting itself three exchanges later.
Anthropic's own testing showed Fable 5's persistent memory tripling its performance retention on long multi-step tasks compared to the previous Opus model. That is a genuine, measurable improvement — and it means a well-designed method that uses Fable 5 as its engine will produce better output than the same method run on a weaker model.
So the model getting smarter is good news. It is just not the news you might assume. A smarter engine does not know your business needs a method. It only knows how to answer whatever question you put in front of it, in whatever order you ask it.
The failure mode nobody talks about
Here is what actually happens when you hand an unguided model — even a brilliant one — an open-ended strategy request: it answers. Every part of it. Immediately.
Ask for a growth strategy, a pricing model, and an org design in the same session, and Fable 5 will hand you all three — confidently, fluently, and completely disconnected from one another. Nothing stops it from pricing a product for a customer segment it has not yet been told to commit to. Nothing stops it from drawing a strategy map before an internal assessment has established what the organization can actually deliver.
A weak model produces an obviously bad strategy. You can spot the gaps because the writing gives it away — thin reasoning, generic language, contradictions in plain sight.
A frontier model produces a great-looking bad strategy. The prose is polished. The frameworks are named correctly. The document reads like something out of a top-tier consulting deck. And underneath the polish, the customer definition on page four still doesn't match the value proposition on page seven, because nothing in the model's design forced those two things to be built from the same foundation.
That is the actual risk of pairing frontier intelligence with no method: it doesn't make bad strategy more obvious. It makes bad strategy harder to catch.
Why the sequence is the real proprietary advantage
TheStrategist.me was never built to compete with a language model on raw intelligence. It was built to solve a different problem entirely: strategy is not one discipline pretending to be twelve. It is genuinely interdisciplinary, and most frameworks — including the best prompt libraries built for the newest models — treat it as a single skill you apply once.
The method runs on four phases I call MoDevExRev — Mobilize, Develop, Execute, Review — and inside those four phases sit five entirely different process groups: Strategic Planning, Value Discovery, Project Management, Change Management, and Leadership. Each one addresses a domain the other four cannot touch. Strategic Planning gives you the analytical core. Value Discovery grounds the whole plan in what a customer will actually pay for. Project Management is the bridge between a plan and a Tuesday morning task list. Change Management accounts for the fact that every strategy asks human beings to work differently than they did yesterday. Leadership recognizes that a strategy is, underneath everything, an act of belief communicated from the top.
Most strategy work — human or AI-assisted — only ever touches the first two. That is why so many strategies read well and still die in the org chart six months later.
Inside those five process groups, the method sequences twelve processes that build on each other in a fixed, non-negotiable order. You do not get to jump to a Value Map before you have prioritized Value Drivers. You do not get to prioritize Value Drivers before you have committed to one Primary Customer. You do not get to commit to a Primary Customer before Internal and External Assessments have told you what is actually true about your organization and its market. Skip a step, and everything built on top of it is guesswork wearing a strategy document's clothing.
That sequence is enforced by 36 carefully curated tools — some borrowed from decades of established strategy literature (VRIO, Porter's Five Forces, the Value Proposition Canvas, OKRs), and some bespoke: diagnostic instruments I built specifically because nothing in the existing literature measured what a specific process step needed measured. A Product/Company Fit Matrix that tells you whether an opportunity actually matches your organization's operating style. A Change Readiness Matrix that scores leadership and organizational appetite for change before you commit resources to an initiative that will die on arrival. These aren't decoration. Each one exists because a specific process, at a specific point in the sequence, needed a specific instrument that didn't already exist off the shelf.
All of that runs through seven distinct expert skills working in concert — the core strategy method itself, plus specialists in customer journey mapping, lean process engineering, risk analysis, business case justification, project execution, and OKR coaching. Each skill is a different lens on the same strategy, applied at the moment the sequence calls for it, so that a risk assessment happens exactly when an initiative needs stress-testing and a business case gets built exactly when a major resource commitment needs justifying — not as an afterthought bolted onto a finished plan.
What actually happens when you multiply a smarter model by the right method
Here is the part that should reframe how you think about this. TheStrategist.me is not in competition with Fable 5. It runs on top of it.
Every one of the twelve processes, every one of the 36 tools, every checkpoint that refuses to let you move forward until the current step is genuinely complete — all of that logic can be executed by whichever model sits underneath it. Run the method on a weaker model, and you get a competent, disciplined strategy. Run the same method on Fable 5, and the disciplined sequence stays exactly as rigorous, while the individual outputs inside each step — the writing, the synthesis, the pattern recognition across a messy set of financials — get sharper.
That is the actual shape of the advantage, and it is durable in a way pure prompt engineering never will be. Prompt libraries get outdated the moment a new model changes how it responds to a given phrasing. A sequencing method does not age that way, because the thing it protects against — a smart system answering the wrong question at the wrong time, in the wrong order — is not a model limitation. It is a structural risk in any unconstrained tool, no matter how capable that tool becomes next year.
As the models get better, the value of the method does not shrink. It grows, because the cost of a fluent, convincing, wrong strategy only goes up as the fluency goes up.
If you want to see the full breakdown — the ten-dimension scoring, the specific failure cases, and where a smarter model genuinely does help versus where it structurally can't — I wrote up the complete comparison here. Send it to the person on your team who just told you they're "just going to ask ChatGPT to build the strategy this quarter." It's a five-minute read that will save them a much longer conversation later.