Why AI threatens generic platforms, not all SaaS

Everest Group published a piece recently called Why Agentic AI Is Breaking The SaaS Pricing Model. It is a sharp read, and the argument lands. Per-seat pricing struggles when one operator with an agent does the work of five.

The pricing question is downstream of a bigger one. When agents can do the configuration work that generic platforms used to charge for, the platforms themselves become harder to justify. Not all of them. Some categories of SaaS are bulletproof. Others are about to spend the next two years pretending they are not already in trouble.

The split is worth thinking about clearly, because the headline that "AI is going to eat SaaS" is wrong, and the defensive line that "AI changes nothing for incumbents" is also wrong. The truth is in between, and it has nothing to do with pricing models.

The boundary-level layer is fine

Stripe is fine. AWS is fine. Twilio is fine. Plaid is fine. The SaaS businesses that sit at the boundary between your application and a regulated, operationally complex external world are not under threat from AI in any meaningful sense.

Why? Because the value they provide is not "we built a clever interface for a thing you could do yourself." The value is that they have done the hard, unglamorous work of integrating with hundreds of banks, dozens of card networks, every major cloud region, every telco, every compliance regime. They carry the operational burden of keeping all of that running. They absorb regulatory risk. They sit inside network effects that compound with every new participant.

An AI agent cannot replace any of that. It can call Stripe's API more efficiently than a human can. It cannot replace Stripe.

This category is the most defensible position in software. Operational complexity, regulatory moats, two-sided network effects. The AI conversation passes right over the top of it.

The middle of the market is in trouble

Now look at the other end. WordPress. Salesforce. HubSpot. Most CMS products. Most no-code and low-code tools. The whole "configurable platform you operate yourself" category.

These platforms share a structural property. They are deliberately schema-less or schema-thin. Their value proposition is flexibility. You can model anything in them, which is the same as saying they do not really know what you are modelling. The platform provides primitives. You provide the structure, the workflow, the data model, the integrations, the rules. The platform charges for the primitives.

That works fine when shaping primitives into something useful is hard. It is the reason these companies built businesses worth billions of pounds. It stops working the moment something else can do the shaping for you.

The dirty secret of no-code

Here is the part the no-code marketing teams will never put on a billboard. The promise of WordPress, Salesforce, and HubSpot was always democratisation. Your marketers can run their own site. Your sales team can build their own pipeline. Your operations people can configure their own workflows. No engineers required.

That promise has not been kept anywhere I have ever seen it deployed. WordPress runs roughly 40% of the web, and it is propped up by a global ecosystem of agencies, freelancers, and plugin specialists. Salesforce is so notoriously hard to operate that "Salesforce administrator" is a salaried profession with its own certifications and a six-figure ceiling. HubSpot's partner directory lists thousands of agencies whose entire business is making HubSpot do what the buyer thought it would do out of the box.

A whole priesthood of consultants exists to operate the platforms that were sold on the basis that you would not need them.

I am not making a moral argument here. The consultants are mostly doing good work, the platforms genuinely are powerful, and the buyers genuinely do get value. The economics are honest about themselves now. The platform sells access to primitives. The consultants sell the labour of turning primitives into outcomes. The buyer pays for both, often without realising that the second invoice is the larger one.

AI collapses the distance between intent and outcome

The reason this arrangement is suddenly fragile is that AI is now competent at the layer the consultants own.

Stop and think about what a buyer actually wants when they purchase a CMS. Nobody wakes up and says "I want a content management system." They want a website that publishes their case studies, gates a couple of lead magnets, integrates with their CRM, sends weekly emails, and ranks for the three terms their sales team cares about. The CMS was always an intermediary. A general-purpose machine that you, or a consultant on your behalf, configured into something specific.

When the configuration step costs months of human labour, the intermediary justifies its existence. When the configuration step is something an agent can do in an afternoon against a clear set of business requirements, the intermediary becomes a cost rather than a service. The buyer does not need a CMS. They need the website. AI removes the layer in between.

This is not theoretical. Every senior engineer I know has watched a Claude or Cursor session spin up a working content site, complete with admin, in a single sitting. The output is not always production-ready. It is often more bespoke and better-suited to the actual requirement than the generic platform alternative. The gap between demo and actually deployed is closing every quarter.

What survives

The SaaS that survives this transition has a property the schema-less platforms lack. An opinionated, structured data model that is genuinely the source of truth for something the buyer cares about.

Think of platforms whose primary asset is the data they hold and the structure they impose on it. A booking system that owns the canonical schedule. A clinical record system that owns the patient timeline. A logistics platform that owns the consignment graph. The website, the mobile app, the partner integration, the BI dashboard, all of these are outputs of that data model. Replace the website tomorrow with an AI-generated alternative and the data model still has to exist somewhere. Whoever owns it owns the customer.

This is the inversion that matters. Generic platforms treated the website or the dashboard as the product, and the data model as a side effect of using the product. Surviving platforms treat the data model as the product, and the website is one of many channels that surface it.

I see the same pattern from the other side every week. Clients sit on top of years of operational data they have never structured properly, never queried meaningfully, and never connected to the decisions the business is actually making. The data is there. The structure is missing. AI will not fix that for them, because AI is not magic and badly-shaped data does not become well-shaped by adding a model on top. The businesses that do invest in shaping their data, the ones that own a real schema for what they do, are the ones that will still have leverage when the generic platforms running on top of that data are cheap and forgettable.

The honest reframe

Everest Group's pricing observation is correct and downstream. Per-seat pricing is breaking because the unit of value is shifting from a person clicking a UI to an outcome produced by an agent, and that is genuinely hard to price per seat.

The bigger story is which categories of SaaS still have a unit of value worth pricing at all. Boundary-level infrastructure does. Structured-data platforms with a real schema do. Schema-less, configuration-heavy generic platforms increasingly do not, because the work they intermediate is collapsing into something AI can do directly.

If you are a buyer of these platforms, the implication is not "rip everything out tomorrow." It is "stop treating your generic platform as your source of truth, and start owning the data model underneath it." If you are running a SaaS business in the middle of the market, the implication is sharper. The configuration moat that funded your last decade is not going to fund your next one. The structured data your customers are accumulating inside your platform is. Find a way to make that the product before the generic wrapper around it stops paying the bills.

The SaaS businesses that will still matter in five years are the ones that figure out which side of this line they are on, and stop pretending the answer is the same for everyone.

If you are working through what this means for your own technology strategy, structuring your data so it survives whichever platform comes next is the kind of question I work on with clients in a fractional CTO capacity. Happy to talk if any of this lands close to a problem you are wrestling with.