Two Arguments About Vertical Software in an Agent World. Both Are Right.

Two recent industry perspectives argue opposite things about where software is heading in an agent world.

In January, Aaron Levie at Box made the case that enterprise software is about to become more valuable, not less, in a world of agents. His argument is straightforward. Agents are non-deterministic. Software is deterministic. The deterministic layer becomes the guardrail that makes the whole system work. When there are a hundred times more agents than people doing knowledge work, systems of record are not commoditized. They are essential. The Total Addressable Market (TAM) for software does not shrink. It expands from a slice of IT spend to a share of the entire cost of human work.

A month later, Gradient Ventures published a piece arguing that public software companies had shed more than twenty percent of their value in two months, and that even companies with traditionally strong moats (such as Salesforce and ServiceNow) were not being spared. Their thesis: a moat alone is no longer enough. The market is repricing the entire category because growth investors are exiting and because agents are quietly hollowing out the workflows that once justified per-seat pricing.

I believe both arguments are correct.

Levie and Gradient are describing two different segments of the software world, and most people are talking about them as if they were the same. They are not. Here is what I think is actually happening.

The Levie argument applies to software that owns something agents cannot easily reconstruct: deep proprietary data, embedded operational workflows, regulatory scaffolding, and decision points where a human still needs to be in the loop. In that segment, agents do not replace the software. They sit on top of it. The software becomes the place where work happens. The agent becomes the way work gets done. Value accrues to whoever owns the substrate.

The Gradient argument applies to software that wrapped publicly available data in a clean interface, charged for the wrapping, and called the wrapping a product! I am intentionally being reductive here to show the contrast. That category is in real trouble. When an agent can reconstruct the workflow on its own, parse the same data sources, and present a result without any of the underlying license fees, the price compresses to the value of the data plus the agent. The license premium evaporates.

A five-question test

I am proposing a five-question test. When there is a software business in question, ask the following:

1. 01What proprietary data does it hold that an agent cannot scrape or synthesize?
2. 02How complex are the workflows it has automated, and how much of that complexity is real?
3. 03How deep are its system integrations, and how many years did those integrations take to build?
4. 04Are there decision points in the workflow where a human is required as a function of regulation, liability, or human judgment?
5. 05Can a third-party agent plug into this software cleanly, or does the architecture make integration hostile?

If the answers are strong, the Levie framework applies. Value is going to expand, and the business will benefit from the agent boom rather than be diminished by it. If the answers are weak, the Gradient framework applies. Repricing is coming, and the question is whether the company restructures fast enough to find a new defensible position before the market does it for them.

What this means, depending on your seat

Depending what role you play today in vertical software, here are the strategic considerations that become clear:

• If you are a product owner or builder, the strategic question is no longer 'what features do we add' but 'what do we own that gets more valuable when an agent shows up.' If the answer is data, double down on data. If the answer is workflow depth, invest in that depth and make it easy for third-party agents to integrate. If you cannot answer the question, that itself is the answer.
• If you are a commercial leader within a business or a buyer of vertical software, the per-seat license is no longer the unit of value. The unit of value is what the software does that an agent cannot do on its own. Pay for that. Stop paying for the rest. The vendors who continue to price by seat in 2027 are telling you something about their confidence in everything else they sell.
• If you are a vertical software investor, the consolidation wave that is starting will accelerate. Companies in the second category will look for shelter inside companies in the first. Some of these potential acquisitions can still be value-creating. You can also argue that many acquisitions will be expensive ways to delay a reckoning.
• If you are a marketer or a P&L owner of vertical software, agents do not care about a software vendor's brand, its history, or its customer relationship. They care about the Application Programming Interface (API) surface and the data they can access. That means the soft moats software has relied on for two decades (stickiness, switching costs, executive relationships, marketing spend) matter less every quarter. The hard moats (proprietary data, complex workflows, deep integrations, regulatory positioning) matter more.
• If you are a GTM (Go to Market) leader or a value engineering leader in vertical software, the companies that come through this in good shape will be the ones that retrain their commercial teams to sell outcomes instead of seats, restructure their product teams to expose data and workflows to third-party agents rather than guard them, and stop measuring success by license growth.

Software is not dying, and software is not invulnerable. It is sorting itself into two piles, and the work for anyone building, buying, or investing in this category is to figure out which pile a given asset is in, and how confident you are in that assessment.

I find this moment more interesting and invigorating than alarming. The companies on the right side of the line have a tailwind I have not seen in twenty years of working in and around software businesses. The companies on the wrong side have a window to do something about it, and the window is closing.

The question for every software leadership team right now is not whether agents are coming. They are here. The question is what the company owns that gets more valuable when they arrive.