The Great Software Repricing: How Agentic AI Is Unbundling SaaS and Redrawing Automation Budgets in 2026

What the $234 billion figure actually means

Gartner's phrase for the mechanism is agentic arbitrage: when an AI agent completes a task across several systems, the person no longer needs to open each of those systems to get the work done. That matters because a large share of a traditional application's value was locked inside its interface. The screens, the dashboards, the click paths that a licensed user navigated every day were both the product and the justification for charging per user. An agent that delivers the outcome directly makes those screens invisible, and in doing so it severs the link between user growth and revenue growth that the whole software-as-a-service model was built on.

Gartner frames the result as a return of the "SaaSpocalypse", the disaggregation of the legacy SaaS market as we have known it. The word sounds apocalyptic, but the forecast is not that software spending collapses. Gartner expects overall software spending to keep growing at roughly 12% a year through 2030. What changes is where the money goes. By 2030 the firm estimates that agentic arbitrage will account for around 20% of enterprise SaaS spend, redirected away from seats and toward outcome-centred services. The pie keeps growing; the slices are being recut.

This sits inside a much larger wave of AI investment. In May 2026, Gartner forecast worldwide AI spending of $2.59 trillion for the year, a 47% jump on 2025. Agentic AI software specifically is projected to reach $985 billion by 2030, growing at a compound annual rate above 60%. The direction is not subtle, and it is not a rounding error in anyone's budget.

Why per-seat pricing is quietly dying

Per-seat pricing rests on one assumption: that the value a company gets from software scales with the number of humans who log in. For two decades that assumption held well enough. Agents break it. When a single agent can carry work that previously occupied several people, seat counts stop being a proxy for value, and a vendor that keeps charging per seat is effectively asking to be paid for logins that no longer happen.

The market has already started to move. Gartner data cited in Deloitte's 2026 technology predictions shows the share of software revenue that comes from seat-based licensing declining from about 21% toward 15% by 2030, while Gartner expects at least 40% of enterprise SaaS spend to shift toward usage-, agent- or outcome-based pricing over the same period. On the vendor side, Kyle Poyar's 2026 State of B2B Monetization survey of more than 230 software companies found hybrid pricing, a base fee blended with usage or outcome components, to be the single most common primary structure. Deloitte's own read is that seat-based licensing "could give way to hybrid approaches that blend usage- and outcome-based pricing", with years of experimentation before any standard settles.

You do not have to be a large enterprise to feel this. The mainstream automation platforms that small and mid-size teams already use are repricing in exactly the same direction, and they are the clearest early signal of where everyone else is heading.

The automation platforms are the leading indicator

Look at the tools most businesses reach for first. In 2026, Zapier folded its AI steps, code and SDK into the same task-based pricing model that governs the rest of the platform, and launched Copilot, an AI builder that assembles a workflow from a plain-English description. Make continues to bill on operations, where each module execution counts as one operation, and pairs that with its own AI agent features. Microsoft's Power Automate splits its pricing between a per-user premium plan and a separate per-bot charge for unattended automation. Each of the three now connects natively to OpenAI, Anthropic and Google models. The common thread is that the meter is moving closer to the work itself: tasks, operations, bots and AI actions, rather than a flat headcount.

Note the last row. The customer-support category has moved furthest toward pure outcome pricing, with vendors such as Zendesk and Intercom charging for a successful AI-driven resolution rather than for an agent seat. Prices in that segment cluster between roughly $0.50 and $2.00 per resolution, with no charge when the agent escalates to a human. It is the clearest live example of the model everyone else is edging toward, and it is worth studying closely because the same contract questions will soon arrive in categories you buy today.

The four pricing models you now have to compare

Because the ground is shifting, a buyer in 2026 has to evaluate not just features but the shape of the bill. Four models dominate, and most real contracts blend two or more of them.

• Usage-based: you pay for what the software consumes — tasks, operations, API calls or tokens. Flexible and fair at low volume, but the bill grows with activity and can surprise you at scale.
• Outcome-based: you pay only when a defined result is delivered, such as a resolved ticket or a booked meeting. Incentives align beautifully, but everything depends on how the outcome is defined.
• Credit-based: you buy a pool of credits that AI actions draw down. Easy to cap, but credits expire and the exchange rate between a credit and real work is often opaque.
• Hybrid: a base platform fee plus usage or outcome surcharges. This is now the dominant structure precisely because it protects the vendor's revenue floor while sharing the upside — which also means you carry both a fixed cost and a variable one.

Each model changes who carries the risk. Per-seat put the risk on you: you paid whether or not the seats were used. Pure usage and outcome models push risk toward the vendor at low volume and back toward you at high volume. The job of a good procurement conversation in 2026 is to find where the crossover sits for your specific numbers, which is exactly the analysis we walk through in our guide to pay-per-resolution, outcome-based AI agent pricing.

What this means for your automation budget

The old budgeting question was simple: how many licences do we need? The new one is harder: how much work will we push through the meter, and what is that worth per unit? Answering it well protects you from the two failure modes of the repricing era — bill shock from an uncapped usage meter, and overpaying a flat rate for capacity you never use.

A practical way to plan a 2026 automation budget:

1. Model realistic volume, not a demo. Estimate the monthly tasks, operations or outcomes you will actually run, then ask the vendor to quote the price at one, two and three times that number.
2. Define the billable unit in writing. If you pay per resolution, the contract has to say what a resolution is. Ambiguity here is where budgets quietly blow out.
3. Pin down the edge cases. Who pays when the agent retries, fails, or hands off to a human? These are the lines that separate a $0.50 outcome from a $2.00 one.
4. Set a hard cap or alert. Usage-based tools should never run without a spend ceiling or a threshold notification. Treat an uncapped meter as an unpriced risk.
5. Compare total cost of ownership, not the sticker. Fold in implementation, integration and internal time before you decide a new model is cheaper.

That last point is where most comparisons go wrong. A per-resolution price looks tiny next to a seat licence until you add the cost of building, integrating and governing the agent, and analysts consistently find that outcome pricing only overtakes per-seat above a few thousand interactions a month once those costs are included. Our breakdown of the total cost of ownership of an AI agent lays out the line items that rarely make it into a vendor's pricing page.

The buyer's advantage hidden in the disruption

There is a genuinely good side to this for buyers, and it is easy to miss under the apocalyptic branding. Outcome pricing, done properly, aligns a vendor's revenue with your results for the first time. Under per-seat licensing, a vendor was paid the same whether the software delivered value or gathered dust, and an automation vendor charging per seat was, oddly, rewarded when its own product under-delivered and you had to keep humans in every seat. Outcome pricing inverts that: the vendor only earns when the agent actually resolves the ticket, books the meeting or collects the invoice. No outcome, no charge.

That alignment is the reason buyers are voting for it. In Futurum's first-half 2026 survey, 43% of buyers said they preferred consumption-based models and 27% favoured outcome-based structures, and vendors that would only sell on a seat-only basis risked being disqualified from deals outright. The leverage has shifted. A buyer in 2026 can reasonably ask a vendor to put some of its fee at risk against a defined result, and increasingly the market answers yes.

The catch is that this leverage only pays off if you are disciplined about definitions and governance. An agent that acts across your CRM, your helpdesk and your finance system to "deliver an outcome" is also touching more of your data and taking more autonomous actions than a seat-bound tool ever did. The pricing conversation and the security conversation are now the same conversation, which is why buying an agent has more in common with a procurement audit than with clicking "start free trial" — a theme we cover in how to buy automation without creating shadow AI.

Old model versus new model, side by side

It helps to see the shift laid out plainly, because almost every assumption in a traditional software budget flips.

Neither column is universally better. A stable, predictable per-seat plan is still the right call for a small team with steady, modest usage that values a number it can defend to finance. The outcome-led model rewards teams with high volume and a clearly measurable result. The mistake is assuming your old budgeting instincts transfer intact. They do not.

A checklist before you sign anything in 2026

If you are renewing a tool or buying a new agent this year, run the contract through these questions before the pricing model, not the feature list, decides the deal for you.

• What exactly is the billable unit? Get the definition of a task, operation, credit or resolution in writing, including what does not count.
• What is the price at peak? Ask for the cost at your expected volume and at three times that, so a busy quarter does not become a budget crisis.
• Who pays for failure? Confirm the charge on retries, errors and human escalations. This is where outcome pricing hides its real cost.
• Is there still a seat charge underneath? Many hybrid deals stack usage on top of seats. Make sure you are not paying twice for the same value.
• Can you cap it? Insist on a spend ceiling or alerts. An uncapped meter connected to an autonomous agent is an open-ended liability.
• What happens to your data if you leave? Portability matters more when the agent, not you, has been touching every system.
• How is the agent governed and logged? An agent that acts across systems widens your audit and security surface; the vendor should make its actions inspectable.

Where this leaves you

The SaaSpocalypse headline is easy to dismiss as analyst theatre, and the $234 billion number will be argued over for years. But the underlying movement is real, measurable and already showing up on the invoices of the automation tools you use every week. Software spending is not shrinking; it is being repriced, from a licence that assumed humans in seats to a meter that tracks work and results. For automation buyers that is neither a crisis nor a windfall. It is a shift in where the risk sits, and the teams that come out ahead will be the ones that treat the pricing model as a first-class part of every decision — modelling real volume, defining the billable unit, capping the meter, and comparing the fully loaded total cost rather than the sticker. Do that, and the repricing works in your favour instead of against you.

FAQ

What is agentic arbitrage in one sentence?

It is Gartner's term for the value that shifts when an AI agent does work across several systems, so users stop opening those systems' interfaces and vendors can no longer tie their revenue to seat counts.

Does the SaaSpocalypse mean software gets cheaper?

No. Total software spending is still forecast to grow around 12% a year through 2030, and worldwide AI spending is set to reach $2.59 trillion in 2026. What changes is the mix, moving from per-seat licences to usage and outcome pricing.

Why is per-seat pricing declining?

Because one agent can do the work of several people, seat counts no longer track value. Gartner data in Deloitte's 2026 predictions shows seat-based revenue share falling from about 21% toward 15% by 2030.

Which pricing model is safest for a small team?

Below a few thousand monthly interactions, a flat or hybrid plan is usually cheaper and far easier to defend to finance. Outcome pricing tends to win only at higher volume, once implementation and support costs are included.

What is the single biggest budgeting risk in the new model?

An uncapped usage meter. Always set a spend ceiling or alert, and pin down exactly what counts as a billable task, operation or resolution before you sign.

How do I know if outcome pricing beats per-seat for me?

Model your realistic volume, add the fully loaded cost of ownership, and find the crossover point. Analysts consistently place it above a few thousand monthly interactions, but your definitions and edge cases move it.

Should I renegotiate an existing per-seat contract?

It is worth asking. Buyer preference has shifted hard toward consumption and outcome models in 2026, and many vendors will now put part of their fee at risk against a defined result rather than lose the deal.