The Mirror Test

I got on a call with a private equity firm a few weeks ago. We were talking through one of my clients, a solid business with a clear fit. It was the sort of target that would normally sit neatly inside a platform story, but I could tell early on that this was not going to be a standard bolt-on conversation.

The shift was not dramatic, which is what made it interesting. There was no big announcement or performative speech about how AI changes everything. Instead, they spoke like a team that had already done the work and quietly updated their internal model of reality.

They told me they had just completed a deep internal review of generative AI. This was not a deck or a high-level discussion. They had their portfolio companies demonstrate what they could actually build by applying these tools into their products and workflows. They spent time getting educated in a “show me what is possible in production” way.

Then they made a decision that would have sounded odd a year ago. They are pausing bolt-ons, at least for the near term, until they understand what their portfolio companies can now build themselves.The logic was simple. If AI has compressed development timelines, why keep paying acquisition multiples for adjacent capabilities? Why buy when those features might now be built internally with a focused team and the right tooling?

For years, bolt-ons worked because adjacency had weight. Even when a buyer technically could build a similar feature set, they did not want the distraction. They avoided the hiring drag, the roadmap trade-offs, and the execution risk. Buying was the shortcut to certainty.

AI has not removed those constraints, but it has changed the speed at which good teams can close capability gaps. “Good enough” arrives faster now. When “good enough” arrives faster, the internal build versus buy conversation becomes more credible.

This firm was clear about what they would still buy. Not features. Foundations.

What they are looking for now are deeper moats. These are assets that sit beneath the application layer and cannot be manufactured quickly with prompts and sprints. They talked about proprietary data that compounds over time. They talked about regulatory and compliance positioning that takes years to earn. They want geographic entrenchment where relationships are genuinely sticky and deep vertical expertise is embedded in the market rather than documented in a wiki.

Those categories have always mattered, but the weighting is changing. AI can help you write code, but it cannot hand you ten years of behavioral data or regulator trust. It cannot build a distribution footprint that was earned relationship by relationship.

I wish this was just one firm with a quirky view. But I have heard versions of this narrative from enough PE shops now that it has become a pattern. This is not an isolated strategy. It is a fundamental shift in how professional buyers value software.

Take the mirror test seriously. Look at your product and ask: What do we have that cannot be replicated by a well-funded team with modern AI tooling and six months of focus?

If the answer is “a lot,” lean into that moat. If the answer is “not much,” you need to think carefully about strategy and timing.