AI MOATS
Where your moat actually is.
Somewhere, a team of twelve is rebuilding your category with no legacy systems, falling model costs, and your margins as their business plan. The board is right to ask where the moat is. The answer is specific: the assets you already own, activated by a loop that compounds them faster than a startup can copy them.
THE THREAT
The disruption math is real.
AI-native startups run the build loop in days, carry no legacy stack, and ride model costs that fall every quarter. Categories with fat margins and slow software are being repriced first. The fear in the boardroom is rational. AI startups took 80 percent of global venture funding in the first quarter of 2026, and the established flank is industrializing too: two of the Big Four are rolling frontier models out to a combined 700,000 consultants. The mistake is concluding that the startup's advantages are the ones that decide the outcome.
THE MOAT
Everyone rents the same intelligence. Your assets decide.
The models defend no one: you, the startup, and your biggest competitor all rent the same intelligence. What decides the outcome is what the intelligence gets to work with. Decades of proprietary domain data, made legible to agents. Trust that took years to earn with customers, parents, patients, and regulators, the asset no startup can fake in high-stakes categories. Distribution that already exists. And the speed of your own loop: how fast you can ship AI changes and prove they behave. A moat in AI is anything the loop compounds that a competitor cannot copy.
STAYING AHEAD
Run the startup's loop on assets they can't have.
The diagnostic produces two maps: where AI-native economics threaten you first, and where your data and trust create leverage a startup cannot match. Then the loop runs on the second map: every slice deposits moat, evals that encode your standards, data made legible, observability that widens, until your company operates with a startup's speed on an incumbent's assets. That is the transformation, framed the way a board actually asks for it. For PE firms, the same diagnostic runs across the portfolio and returns an exposure map and a moat map per company.
COMMON QUESTIONS
Asked by boards, answered plainly.
Is our data actually a moat?
Only once it is activated. Raw data sitting in silos defends nothing: a startup ships around it. Made legible to agents, governed by evals, and compounding through the loop, your decades of domain truth become the one asset a competitor cannot synthesize at any price.
Should we build our own models to stay ahead?
Sometimes, and the evals tell you when. Models are increasingly swappable; the durable assets are the standard that defines what working means for your business and the data underneath it. Where a custom model wins, evals prove it before you commit. See /own-your-models and /custom-ml.
How fast could an AI-native startup actually catch us?
Honestly: months, for features. Years, for what you already own: customer trust, regulatory standing, distribution, and proprietary domain data. Those only defend you if they are activated while the window is open. The window is now.
We're a PE firm. How does this work across a portfolio?
The same diagnostic runs per portfolio company and produces two maps: where AI-native economics threaten each portco first, and where its data and trust create leverage a startup cannot match. One reliability standard then compounds across the portfolio. See /who-we-work-with/pe-portfolios.
LET'S TALK
Bring us the hard problem.
We'll bring the team that ships.