Key Takeaways

  • Naveen Rao warns that venture capitalists backing pure software SaaS face vanishing margins as value flows directly to underlying model providers.
  • The easy software distribution that fragmented enterprise tooling over the last two decades will not support standalone software markets twenty years from now.
  • Coding agents like Cursor are overrated as long-term standalone markets because foundation model providers will absorb their core utility.
  • Software development will experience Jevons Paradox: cheaper code generation drives higher aggregate demand for software engineering and system maintenance.
  • The defensible business model shifts to tech-enabled work combining proprietary automation with human verification, following the pattern set by legal platform Harvey.

The Intelligence Layer Is Eating Pure SaaS

For twenty years, enterprise software followed a predictable playbook. You built a clean interface on top of a database, charged thirty dollars per seat every month, and enjoyed 80 percent gross margins. Easy internet distribution created thousands of fragmented software categories.

Rao built Nervana Systems, sold it to Intel, ran MosaicML, and sold that to Databricks. He watches infrastructure economics closely. In his view, the per-seat software delivery model is dying. When an application's primary value comes from inference rather than workflow logic, the economics invert.

“I would say if you're a pure software investor, you're kind of [screwed] honestly,” Rao says. “And I'll say that pretty confidently because SaaS I don't think is going to be where you gain margin anymore.”

When a thin interface sits on top of a large language model, the company providing the interface pays high token bills while competing with a dozen identical wrappers. The margin pools split between two players: the hardware and model providers running compute, and the companies selling completed work directly to end customers.

Why Coding Agents Shift to Model Providers

The most visible software category today is agentic coding. Tools like Cursor captured immense developer mindshare and delivered real speedups. Rao argues that treating coding tools as independent platforms misunderstands where the defensive barrier sits.

“I actually think that agentic coding stuff is quite overrated in terms of a market because I think it moves to the model provider,” Rao explains. “I don't think you get much value from the others. I know Cursor had a great exit and that's wonderful. But the model providers, the model itself is where most of the value is coming from.”

If the reasoning capability lives inside the frontier model, the model provider can capture the developer workflow by exposing native code completion directly into developer environments. A workflow layer that does not own proprietary models or physical compute sits on shaky ground.

Cheaper Code Triggers Jevons Paradox

Many tech founders assume that automated code generation means companies will spend less money on software engineering. Rao argues the opposite. When the unit cost of producing code drops, aggregate demand for software spikes.

“It's not that everyone's job goes away because the same task could be done for less with AI plus humans,” Rao says. “It's that now I yes that is true, but I will now do way more of those tasks because the demand side is increased. When the supply gets cheaper, it changes the demand side as well.”

This is Jevons Paradox in practice. Cheaper creation leads to more deployed code, more complex integrations, and larger surface areas requiring rigorous maintenance. The winners will not be tools selling code snippets. The winners will be firms that package automation with human oversight to guarantee correct technical execution.

“I'm seeing a crop of companies come where they're taking engineering tasks that have existed for a long time that were done in a traditional way and have to be done correctly,” Rao notes. “And they're making those tasks much faster, much cheaper by using AI plus humans. It's great. And that's, I mean I would say Harvey was a first example of this.”

What to Do With This

Audit your product roadmap and pricing page before your next board meeting. If your software charges per seat for access to a workflow canvas or dashboard, redesign the tier around verified work outcomes delivered by AI with human sign-off. Stop selling software access; sell completed, guaranteed work units priced against the human labor you displace.