Key Takeaways

  • Devin Mathews published a 10,000-word piece in March titled “The Ghost of Software Future” arguing enterprise software will domesticate AI, not collapse under it.
  • Ken Fine estimates that building software represents only 20% to 25% of total resource costs, with 75% to 80% locked up in ongoing maintenance, entity resolution, and integrations.
  • Private equity investors who try to build bespoke internal tools contradict their own underwriting thesis that specialized software creates defensible enterprise value.
  • Many buyout shops built flashy internal prototypes over the past year, but remain 10% away from closing the loop on real production utility.

The Fallacy of the SaaSpocalypse

The market spent the last two years predicting a total collapse of enterprise software pricing power. The bear case argued that cheap foundation models and automated code generation would wipe out classic systems of record. Anyone with an API key could spin up a custom tool, cutting out third-party enterprise vendors entirely.

Mathews rejects that thesis directly. “We wrote a piece in March called the ghost of software future,” Mathews noted. “It was basically like no, enterprise software will domesticate AI. AI is software.”

Incumbents hold the true defensible asset: the proprietary context graph, entity resolution pipelines, and years of historical transaction records. AI models do not replace this infrastructure. They sit on top of it. A raw language model has no institutional memory; enterprise platforms provide the structured environment that makes models usable for deal sourcing and pipeline management.

The Maintenance Trap of Vibe Coding

Cheap engineering tools tempt investment committees into believing they can build their own internal deal engines. Non-technical partners throw prompts at coding assistants, generate workable interfaces over a weekend, and declare victory over expensive SaaS contracts. Fine points out that this enthusiasm ignores the real economics of software engineering.

“There's a saying generally in software that when you build software, and I think this is even more true with AI, that you put roughly let's say a quarter to a fifth of your resource into building it and then the rest into maintaining it,” Fine said.

Initial builds are cheap, but the hidden cost lives in schema drift, edge cases, permission tiers, and ongoing API maintenance. Fine observed: “The friction of getting started is very low. Therefore, you can try a lot of stuff, which is great. But then you have sprawl that you need to manage or deprecate.”

Mathews also highlighted the intellectual contradiction for private market investors trying to become software developers. “If you're a software investor, why roll your own? You're kind of nuking your own investment thesis that software can just you could just vibe code it yourself.”

Prototypes are easy to spin up, but closing operational gaps requires full product teams. As Mathews observed about recent fund experiments: “I'm convinced that there are a lot of funds out there that have built some really cool stuff in the last few months. And it hasn't fully closed the loop yet. They're still 10% away from actually closing the loop.”

Why It Matters

Capital allocators who confuse low barrier to entry with sustainable total cost of ownership are building legacy tech debt inside their own management companies. Markets reward specialization, and firms that attempt to engineer custom internal platforms dilute their core edge in deal execution while underestimating the ongoing maintenance burden of live data systems.