The Repricing of Enterprise Software

There's a question I keep coming back to that I don't think the industry has a clean answer to yet: in a world where agents are doing the work, how do you price the software they work through?

Let me try to think through it.

Most enterprise software we've built over the last few decades has been a system of record. Salesforce holds your customer relationships. Snowflake holds your data. Excel holds your models. These weren't just storage — they were the primary interface between your business and its information, and the humans navigating them were the unit of value. So pricing by seat made sense. You scaled headcount, you scaled seats, the model worked.

Agents break that assumption. Not because systems of record go away — enterprises still want secure, governed, structured environments, and I think that preference is actually going to intensify as agents get write access to sensitive systems. But the human seat stops being the natural unit of value when an agent can navigate these systems directly. The system of record gets demoted. Not eliminated — demoted. From engine to plumbing.

Here's where I want to be careful though, because plumbing isn't automatically low-margin. AWS is plumbing and it's one of the most profitable businesses ever built. The risk for incumbent SaaS isn't just that they become infrastructure — it's that they become interchangeable infrastructure. The margin compression comes from substitutability. If an agent can be pointed at Salesforce or HubSpot or a custom CRM with roughly equal ease, the switching costs that justified premium pricing start to erode. That's the real threat, and it's worth naming precisely.

So who actually forces the repricing? I don't think it's new entrants alone, and I don't think Salesforce or Snowflake voluntarily walk away from seat-based models. The forcing function is the enterprise buyer. Think about what happens when a CFO looks at a 500-seat Salesforce contract and realizes their agents don't need seats — they need API access. That's a procurement conversation that's already starting to happen in some companies. Buyers will drive this repricing faster than competitive pressure does, because buyers have contracts up for renewal and real leverage.

That raises the harder question: what does the alternative actually look like? My intuition is work-based or workflow-based pricing — you pay for what gets done, not for who's doing it. For the agent layer specifically, token-based pricing with some markup is probably the natural model. But that's a different problem than how systems of record get repriced. If an agent touches Salesforce 10,000 times a day, does Salesforce charge per API call? Per workflow completed? Per business outcome? There's no settled answer here, and I think that ambiguity is where a lot of the strategic uncertainty lives right now.

The full-stack question sits underneath all of this. Most companies today operate cleanly in one world — they're either building the system of record or building the agent layer. Very few are doing both, and the ones trying find it genuinely hard. Part of that is product architecture — agents need read and write access to systems that were designed around human interaction patterns. Part of it is trust — giving an agent write access to your CRM or your financial data is a different risk profile than giving a human a seat. And part of it is go-to-market: how do you sell a product that serves both a human workflow and an automated one without cannibalizing yourself?

The companies that figure out that last question — not just the product, not just the architecture, but the pricing design — are probably the ones that define what enterprise software looks like in five years. Right now, most of the conversation is about what agents can do. I think the more durable question is how we charge for it.

I don't have a clean answer. But I think it's the right thing to be sitting with.