Amie Devero:
During 2024, what I have come to think of as “Year of the LLM”, I tried vibe coding. I wanted to make a gardening app. But Francis, my friend and former business partner, had just led a strategy retreat and had his own AI challenge. His client needed to share a new strategy with a very conservative Board of Directors. PowerPoint would not be enough.
Francis had been trying to achieve it with NotebookLM — but it was proving too blunt an instrument. He asked if I would try to vibe code something.
Immediately I hit a wall. I gave the documents to a vibe-coding LLM and asked it to produce an interactive tool to explain the strategy. What it produced was okay but needed tweaking, so I tried again. The next time, it seemed to be talking about a completely different strategy. Familiar, but not the same.
By then I had worked with AI enough to understand the problem. Each time I uploaded the documents, the LLM interpreted them differently. Documents only deliver a facsimile — and everyone reads it differently.
What I needed was a single articulation of the whole strategy, in a form that could go from test to test, and ultimately, from director to director. It couldn’t be just nuts and bolts. It needed nuance, rationale, texture. After many failures, a “schema” emerged.
It included the choices, the tradeoffs, what the organization had decided not to do, how it all connected and perhaps most importantly, why.
It was the full strategy, expressed as data.
Now I could ask an LLM to use it over and over, knowing it would understand the same thing each time.
That made it possible to build the tool Francis’ client needed.
But the question of how to share a strategy was now an ear-worm I couldn’t turn off. For example, Alice. As with many of my clients, I was her coach but had also led her organization’s strategy retreat.
Within weeks of the retreat, the strategy had vanished from our conversations. Her concerns regressed. Soon, we were once again discussing her customer success team and how they kept coddling customers and doing their configurations — the exact thing
the strategy killed when it pivoted to a self-service product.
There it was again, back from the dead like Lazarus.
What had happened?
A struggling client called and the rep fell into the habit of taking over. He didn’t know how to do anything else. Sure, self-service would be great. But what should he do with this customer, right now?
Like the LLM, each person picked out bits of the strategy to embrace, and neglected others.
The retreat was successful, and the strategy it produced was brilliant. But it didn’t exist in a form that could easily map onto anxious client calls or the pressure to close a deal.
Nothing connected it to the work.
Each business function communicates with fixed artifacts. Product has a roadmap.
Finance has the ledger. Sales has a CRM. And those artifacts do more than
communicate. They are what everything else gets built on. Nobody forecasts from a memo about the finances. They forecast from the ledger.
Strategy has… slides?
Yes, there are metrics. But those measure strategic outcomes, not the choices or the reasoning behind them.
A schema gives strategy its own artifact. Any AI skill can be applied to it.
And when the strategy adapts, the schema adapts with it.
Eighteen months later, Contextuum — the company I built around that schema — uses it to generate on-demand, job-specific strategic education for teams. That’s just the most obvious use of a schema. But it’s not the most interesting one.
I still haven’t finished the gardening app.
But I may have an answer to the question Alice’s team provoked: What would it take for your strategy to reach to every Tuesday afternoon decision?
