Gartner Says 35% of Point SaaS Gets Absorbed by Agents. Start the Clock

Gartner projects that roughly 35% of point-product SaaS tools will be replaced or absorbed into agent ecosystems by 2030. That is a third of a category disappearing into someone else's platform over five years. The projection is widely quoted and rarely examined, and the interesting question is not whether it happens but which third.

What the projection actually says

The claim is narrower than the way it gets quoted, and the narrowing matters.

It concerns point products, meaning tools that do one job well and are bought separately from the systems they connect to. It does not cover platforms, systems of record, or software carrying regulatory obligations.

It says replaced or absorbed, which are different fates. Replacement means a competitor wins the deal. Absorption means the function stops being a separate purchase because a platform the customer already pays for now does it.

And it runs to 2030, which is a five-year window. That is a strategic horizon, not a quarterly one, and the difference decides how urgently anyone should act.

How absorption actually happens

Nobody wakes up and cancels a working tool. Absorption follows a sequence, and the sequence is slow enough to watch.

The critical detail is stage 2. Platform features rarely match a specialist product. They reach the point where they are adequate for the majority of use cases, and adequate plus already-paid-for beats better plus separately-invoiced in most procurement processes.

That is why product quality is a poor predictor of survival here. The question is not whether you are better. It is whether the gap between you and adequate is large enough that someone will keep writing a separate cheque for it.

Which categories are structurally exposed

Exposure follows the shape of the work rather than the size of the vendor.

The top of that chart is uncomfortable reading for a large number of profitable companies. Moving data between two systems, alerting someone when a condition is met, and presenting information in a readable form are all things a competent agent does without a dedicated product.

The bottom of the chart is where the durable businesses sit, and the reason is not sophistication. It is obligation. A system that holds the authoritative record of what happened, and can evidence it to an auditor, is carrying a responsibility that nobody wants to transfer to a general-purpose tool.

What actually protects a category

Four properties, in order of how reliably they hold.

Regulatory obligation. If replacing your product requires a compliance review, a re-certification or an auditor sign-off, the switching cost is measured in months and budget rather than in effort. This is the strongest protection available and it is available to fewer companies than claim it.

Authoritative data. Not volume of data, which is common, but data that is the source of truth for something. If your customer's answer to "what actually happened" comes from your system, you are not a point product.

Domain logic accumulated over years. Industry-specific rules, edge cases and validation logic that took a decade of customer feedback to encode. A platform can build this and the return on doing so is poor relative to its alternatives.

Multi-party network position. If switching requires several organisations to move together, no single customer can leave unilaterally. This is rare and extremely durable where it exists.

Notably absent from that list: user experience, model quality and feature depth. All three matter for winning deals and none of them survives a platform deciding to compete.

Why 2030 and not 2027

Enterprise friction sets the clock speed

The five-year window is doing more work in this projection than the percentage is.

Enterprise switching is slow for reasons that have nothing to do with AI. Contracts run for years. Security review takes months. Data migration is unglamorous and expensive. Change management fails more often than it succeeds.

Those frictions do not disappear because a better option exists. They set the clock speed of the entire process, and they are why a projection about 2030 says almost nothing about what happens to a software company's revenue next year.

What the window is actually for

The practical implication for an operator is that you have time and you do not have unlimited time. The strategic choices that determine survival need making while the revenue is still there to fund them, which is the argument set out in the kill zone test.

What the evidence currently supports

The direction is well supported. The scale is a projection and should be read as one.

Evidence on both sides

Supporting the case: enterprises are actively rationalising software spend, coding agents have collapsed the cost of internal builds, and several categories have already been absorbed into platform defaults with no dramatic event marking it.

Cutting against it: the average enterprise ran 291 SaaS applications in 2025, up from 110 in 2020. Application counts are still rising, not falling. Companies are cancelling and buying simultaneously, which is what rationalisation looks like rather than collapse.

A third fact sits between them. Categories absorbed so far have generally been ones where the platform gained something by absorbing them, either data, engagement or a reason to raise its own price. Absorption is a commercial decision by the platform, not an inevitability of the technology.

Both facts are true and they are not in tension. Consolidation within categories can coexist with expansion in the number of categories, and that is the most likely shape of the next five years. The market repricing that assumed otherwise is examined in the analysis of the software selloff.

What this means for a buyer rather than a vendor

If you are the one purchasing software, this projection is an argument for shorter contracts rather than for cancelling anything.

A three-year commitment to a point product in an exposed category is a bet that the category still exists in its current form when the term ends. A one-year term with renewal rights costs slightly more per year and preserves the option to consolidate when a platform catches up.

The second implication is about data portability. If a tool in an exposed category holds data you would need to move, negotiate an extraction clause now rather than at the point you want to leave. Vendors grant these far more readily during a sales process than during a cancellation.

Where this argument is weak

Three problems worth naming.

It is a projection from a research firm, not an observation. Analyst projections about software categories have a mixed record, and this one has not yet been tested against any measurable outcome.

The definition of a point product is not crisp. Almost every software company would argue it is a platform, and the boundary between the two is exactly where the projection's accuracy is decided.

And absorption assumes platforms want the work. Platform companies have finite roadmap capacity and they prioritise what grows their own revenue. A category that is unprofitable, heavily regulated or operationally messy can survive indefinitely simply because nobody with the power to absorb it wants to.

That last point is underrated as a defensive strategy. Being in a business a platform finds unattractive is a weaker moat than owning data, and it is a real one, and it has protected a great deal of software for a long time.

Frequently asked questions

Where to start this week

One question, answered honestly about your own product or your own stack.

If the software were switched off for a week, would a workflow get slower, or would the business be unable to evidence what it did? The first answer describes a coordination tool and it sits in the exposed half of that chart. The second describes a system of record and it does not.

Most software companies believe they are in the second group. Most are in the first, and the distinction shows up in churn data long before it shows up in a market capitalisation.

If the answer is uncomfortable, the useful next step is not panic. It is picking one of the four protections in the table above and building toward it deliberately, while the revenue is still there to fund the work.

Related reading