AI Won't Lift Software Evenly. It Will Split It in Two.

The consensus on AI and software is a rising-tide story: models get better, every application gets more valuable, the whole category floats up together. That story is half right, and the wrong half is the expensive one. AI doesn't lift software uniformly. It bifurcates it: lifting judgment and flooding tooling. The gap between those two halves is where I believe the next repricing happens.

Start with what actually changed. For most of software's history, the scarce, expensive thing was building the tool. A method (how a practitioner underwrites a loan, weights a rubric, sequences a diagnosis) could only reach scale by convincing a software company to build a generic version of it, at which point the method stopped being the practitioner's. My mentor, Vincent Hunt, calls the alternative now emerging method as software. A practice rendered executable by the person who holds it, without a budget or a team standing between the two. What had made this impossible was the cost of building—AI has now collapsed that cost.

But I think there's another consequence people aren't talking about.

If a practitioner can rebuild a generic tool in a weekend, then software whose moat was being the tool loses its pricing power. Commodity, seat-priced, horizontal workflow software (the layer that charged rent for being the only convenient place to do a thing) is exactly the layer AI makes cheap to reproduce. When software stops being scarce, it stops being the advantage. The tooling floods, and flooded things get cheap.

So where does the value go? Not into the tooling, which becomes easier and cheaper to reproduce, but into the one thing abundance makes scarce: trust. When anyone can ship an instrument, instruments stop being the scarce good and trust is what becomes it. The scarce question is no longer "can you build the software" but "whose judgment is inside this one, and does it hold." That is a verification question, and verification is a business.

This is the part the market hasn't finished repricing. The durable winners of an AI-flooded software world are the provenance businesses. These are the ones whose entire product is a trusted stamp that says this method is sound. That footprint already exists in public markets: the ratings agencies, the index and financial-data providers, the risk-scoring and analytics firms whose value isn't a feature set but the credibility of their judgment. Moody's and S&P Global sell essentially one thing, which is the market's trust that their assessment holds. MSCI sells verified methodology as infrastructure. Fair Isaac sells a score the whole lending system is built on.

What's interesting is that I don't think the market has fully recognized this shift yet. Many of these companies are still being evaluated like traditional software businesses, even though their real value comes from trusted judgment. These aren't the companies most people associate with AI, but I do think they're some of its biggest beneficiaries. As AI makes it easier to create software, the need for trusted judgment only grows. The more choices AI creates, the more valuable it becomes to know which ones deserve our trust.

I could be wrong about the timing, or even about whether people will care at all. This thesis falls apart if trust turns out to matter less than I think it does. If businesses simply choose whatever tool is free or good enough, without asking whose judgment shaped it, then software becomes just another commodity and the value never shifts. I don't think that's where we're headed. The higher the stakes, the more people want to know where the thinking came from and whose judgment they're relying on.

That's the bet I'm making.