This year I have worked with PE-backed vertical SaaS companies on how to respond to AI agents. I had seen vertical software up close before, at startups and enterprises in education and construction, where knowing an industry well opens doors and also makes every customer slower to win. Most of this year's conversations come back to one number, and then to a harder question than the number suggests.
The exit math changed
At the multiples reported for middle-market technology deals, a company acquired in 2023 at around 10.2 times EBITDA and sold in the first half of 2026 at 5.7 times would need nearly 80% more EBITDA to reach the same enterprise value. That simplified calculation sets aside transaction costs, debt paydown, and differences between individual companies, but it shows the scale of the problem facing owners who bought near the peak.
Vertical SaaS companies in PE portfolios sit in the technology deal market where those multiples fell. Most are profitable and most have loyal customers. The question for anyone running one is which parts of the business still earn a premium, and how to build more of them before the next exit.
Keeping the customer, losing the workflow
The central risk for a vertical SaaS company is easy to miss because nothing obvious would break. Customers would keep renewing, the database would keep growing, and churn would stay low, while the work people used to do inside the application moved to an agent on top of it.
What is observable today is that enterprises are building agent interfaces on top of their systems and replacing selected point tools. The agent reads the vendor's data, runs the workflow, and answers the user's question, so the user has less reason to open the application. The financial consequence is still a risk rather than a measured trend, and as the previous piece noted, no broad data set yet shows seat counts falling across many enterprises. If agents reduce the number of paid seats or take over billable workflows, the vendor keeps the customer and the data while the value of the workflow, which is what the seat-based subscription was really charging for, moves to whoever owns the agent layer. That could be a model lab, an AI-native vendor, or the customer's own engineers. The renewal would then shrink even though the customer never leaves.
Here is this piece's version of the series question. When software gets cheap, does the vertical SaaS vendor keep the value of the workflow, or only the record underneath it?
Two timelines
In the near term, the squeeze may protect incumbents. As the startup piece in this series describes, seed investors are funding fewer conventional software companies, so fewer funded challengers will attack an entrenched system of record. Leaving one is often a multi-year project, and the enterprise examples in the previous piece show customers rebuilding point tools while keeping their core systems.
In the longer term, the pressure comes from the customer's side, as agents take over the workflow on top.
The strongest objection to this argument is that AI may strengthen established vendors instead. Agents need trustworthy data and rules to act on, and the system of record holds them. Gartner's Patrick Connaughton made this point in an October 2026 webinar, arguing that putting AI interfaces on top of a system of record can deepen the customer's dependence on it. Both outcomes are plausible. Which one a given vendor gets depends on what it controls and how it prices access to it, which is what the rest of this piece covers.
Where defensibility comes from now
The market's verdict on the category is visible in listed business software, where vertical SaaS traded at a median of 2.5 times revenue in August 2026 against 4.7 times for generative AI companies. Within the category, defensibility comes from four dimensions. They are not alternative strategies, and a strong vendor can hold all four.
Each dimension also carries two kinds of value. Switching costs keep customers from leaving. Lasting advantages give customers reasons to choose the vendor again. The difference matters now, because cheap code, reverse engineering, and agents that move data between systems are making switching cheaper, while lasting advantages hold up better.
Dimension | Switching cost (weakening) | Lasting advantage (holding) |
|---|---|---|
Data and domain authority | Historical records that are hard to migrate | Domain rules and context that help agents act correctly |
Platform ecosystem | Integrations a customer has already built | Developers and agent compatibility that attract new customers |
Trust and accountability | Compliance work a customer would have to redo | Audit trails and reliability buyers now require |
Deployment sovereignty | Residency setups that are costly to move | Meeting local rules that rivals cannot easily meet |
A vendor that relies on the left-hand column is relying on something that is getting weaker. Investment belongs in the right-hand column.
An example from construction management
Construction management software shows where the line could fall. Much of the paperwork that fills a project manager's day is a natural fit for agents. Drafting requests for information, logging submittals, writing daily reports, and turning field notes into schedule updates are repetitive, text-heavy tasks that an agent can do from photos, emails, and site notes. A vendor that charges per seat for those screens is exposed. The record underneath is harder to replace. The contract and every approved change order, the payment applications and lien waivers that decide who gets paid, and the project history both sides turn to when a dispute reaches a lawyer all carry legal and financial weight. Owners, lenders, and courts rely on that record, and it has to be right. A construction vendor that holds it, and charges for the validated changes agents make to it, keeps its place even if nobody opens its daily-log screen again.
What to do, and when it applies
The moves below are conditional. Each one helps under some circumstances and hurts under others.
Expose meaning, when the meaning is the vendor's own. Publish the domain's entities, state changes, and rules as APIs, so customer agents act on business context instead of raw tables. An API over proprietary history and trusted domain rules makes the vendor the place agents have to go. An API over commodity features makes the vendor easy to replace.
Price for agents, when customers are already using them. Move from seats to actions or outcomes, with cost limits a CFO can sign. In practice, outcome pricing today usually means tasks completed and agent utilization, because CFOs rarely agree to share value they cannot measure.
Offer a paid on-ramp, for complex, high-value customers. Starting a customer with a paid forward-deployed engineer can deepen the relationship, but it makes the vendor more service-intensive, and that does not automatically help its valuation. Services revenue usually earns lower margins than subscriptions and lower multiples at exit. The play works best when the engagement is time-limited and converts into subscription use, and when each engagement adds reusable specifications and templates, so the tenth customer costs less to onboard than the first. Without those conditions, the vendor may build a stronger business with different unit economics, which can still be a good business but will be valued as one. The startup piece in this series makes a related point about companies that are profitable without fitting the profile that earns the highest multiples.
Open in tiers, as governance allows. Not all access carries the same risk. Read access, which lets agents retrieve documented business context with authentication and access controls, can come first. Write access, which lets agents change records, needs explicit permissions and logging, with safeguards such as validation and rollback. Autonomous action, such as initiating payments or submitting regulated filings, should wait until the platform supports identity for each agent, authorization, approval rules, and an audit trail. A company that allows autonomous action before it can govern it takes on liability it cannot measure.
Sell sovereignty, where customers face residency rules or pay for local control. Offer data residency and audit records that customers can export.
Underneath all of these sits speed. A vendor that writes its domain's meaning into specifications its own coding agents can build from moves faster than a customer attempting a rebuild.
The defensive move that backfires
As agents take over the interface, some vendors will be tempted to protect their economics by restricting or charging for access to customers' own data. Gartner describes a scenario in which that move becomes the tipping point that sends customers to AI-native vendors and custom builds.
Hiding the platform also protects less than it once did. Tools such as REA (Reverse Engineer Anything), an open-source project that connects coding agents to reverse-engineering tools, can study a vendor's web application and recorded network traffic to work out the shape of an undocumented API. A customer or competitor can learn much of what a closed platform tries to keep private.
Opening up is not automatically the answer either. Before opening an API, a vendor needs to know what sits behind it. If it is proprietary history, trusted domain rules, regulated transactions, or the current state of a customer's workflow, opening it can make the vendor more necessary. If it is functionality a rival can match, opening it speeds up commoditization.
Being necessary is also different from being paid. A system of record can stay indispensable to every agent a customer runs and still lose the economics its application used to earn. Buyers will treat access to their own data as a contract term, and the procurement checklist in the previous piece asks them to fix its price for the length of the agreement. Access billed per call can bring in less than the seats it replaces. A vendor that charges only for retrieving records is selling a commodity. The pricing power sits in what it adds on top of the record, such as domain rules, validated transactions, and an audit trail a regulator will accept.
Questions vertical SaaS companies need to answer
If customers stopped opening the application tomorrow, which revenue would remain?
Which parts of the platform would make the company more necessary if opened to agents, and which would make it easier to replace?
If a customer's agent wrote to the platform tomorrow, could the company show what it changed and why?
Is the company optimizing for its exit multiple or for durable cash flow, and does its mix of services and subscriptions reflect that choice?
Read next in When Software Gets Cheap
The Squeeze on Software Startups (Startups). The next piece looks at the seed stage, where the companies that once would have grown into this middle are finding it hard to raise venture capital, and where cheap software may produce a different kind of company.
Also in the series:
What Got Repriced (The Market): Why profitable software companies lost their valuation premium, and who keeps the surplus when code gets cheap.
What the Corporate Buyer Decides (Enterprise): How corporate buyers weigh economics, control, and accountability when deciding what to build, buy, and renegotiate.
Who Keeps the Value When Software Gets Cheap (Closing): The answers drawn together, and the signals worth watching.
Selling the Finished Job (Services): When software gets cheap, does the money move to the work the software does?
Sources
Middle Market Transaction Update, Fall 2026, Mercer Capital (GF Data)
The tech valuation advantage has flipped, Randy Woods
SaaSpocalypse analysis, Dealroom, updated September 11, 2026
The Future of Enterprise Applications: Navigating AI, SaaS Disruption and Vendor Lock-In, Gartner webinar with Patrick Connaughton, October 9, 2026
REA: Reverse Engineer Anything, GitHub