Francis Frangipane The Three Battlegrounds

I first came across the Three Battlegrounds framework when someone linked it in a Discord thread about AI startup positioning. I was skeptical at first — another consulting-style model, right? — but it actually held up when I tried to use it to evaluate a few projects. The core idea is straightforward: in the current AI landscape, there are three main arenas where competitive advantage is being won or lost, and most people who say they're competing are only really playing in one of them without realizing it. Here is what the framework covers. The first battleground is infrastructure. This means whoever controls or has best access to compute, chip supply, and the underlying model layer. The second is data. Not just any data — proprietary, high-quality, domain-specific datasets that fine-tuning or RLHF can actually move on. The third is distribution. This is the go-to-market layer: user reach, platform integrations, and ecosystem lock-in. The reason this triad matters is that most founders and even a lot of investors conflate two of them and think they have a moat when they don't. I ran into this exact problem last year when advising a team building an AI-powered legal research tool. They had strong data assets from a partnership with a mid-size law firm and they were convinced that was their competitive advantage. It turned out their biggest risk wasn't data quality — it was that the compute cost per query was eating their margin because they hadn't locked in infrastructure pricing. They pivoted to using smaller fine-tuned models on cached infrastructure and the unit economics worked. The framework forced them to look at all three battlegrounds instead of fixating on one.

How to Actually Use This Framework

The way I've found it useful isn't as a diagnosis tool. It's better as a stress test. Pick your project or company and for each of the three battlegrounds, write down what you actually control versus what you depend on. Be honest about dependencies. Most people discover they have zero real control over one of the three and it's usually the one they thought was their strength. Here is the part nobody talks about enough. The framework assumes the three battlegrounds carry roughly equal weight. In practice they don't. If you're building a vertical AI application on top of open models, infrastructure is largely someone else's problem — you are renting it. That means your real battlegrounds compress to data and distribution, and the framework effectively becomes a two-question test. If you have neither proprietary data nor a distribution channel, you are building a wrapper and the model authors can replicate you in a sprint cycle. I hit an edge case with this in a project where the data was genuinely proprietary but the distribution required partnering with an incumbent platform. The platform had every incentive to absorb the feature and bury it. The workaround was to structure the deal so that distribution rights included a revenue share that made it economically painful for them to replicate internally. It is not a perfect solution but it bought us eighteen months of head start, which is often enough in this space.

Pitfalls and Where It Falls Apart

The framework has real limitations. It was designed as a strategic lens, not an operational playbook. It does not tell you how to win in any of the three battlegrounds — only that competition is happening there. If you need tactical guidance, you will have to bring your own. Another issue is that the landscape shifts fast. Compute barriers lower every quarter as open models improve and inference optimization catches up. Data moats erode as scraping and synthetic data generation become cheaper. Distribution advantages can vanish overnight when a major platform changes its API policy. I would also note that the framework is less useful for well-funded incumbents. If you already have scale in one battleground, the model does not help you much with the conversion math into the other two. It is most valuable for smaller teams trying to figure out where they actually have a shot. For those teams, the honest answer is usually to pick one battleground and go deep, not to try to compete across all three simultaneously. That last point is where most failures happen — spreading thin across infrastructure, data, and distribution with no real advantage in any of them.