SaaS Challengers: Why AI-Native Startups Are Coming for Every Incumbent

For two decades the safest bet in technology was that an entrenched SaaS incumbent could not be dislodged. Switching costs were brutal, integrations were sticky, and the data was locked in. That assumption is now being openly challenged. Y Combinator's Summer 2026 Request for Startups names "SaaS Challengers" as a category it wants to fund: AI-native replacements for incumbent software, built with better economics and workflows rethought from scratch. And the prediction underneath it is striking — a16z partner Angela Strange expects that in 2026, major financial institutions will start letting legacy vendor contracts expire and replace them with AI-built alternatives, because the risk of not modernizing will outweigh the risk of switching.

The Risk of Not Modernizing Just Flipped

The reason incumbents felt safe was a simple asymmetry of risk. Replacing a core system was dangerous — migrations failed, data got corrupted, workflows broke — so the rational choice for a risk-averse enterprise was almost always to stay. Inertia was the safe option. What Angela Strange is arguing is that this asymmetry has inverted. When AI-native alternatives can deliver dramatically better economics and capabilities, the cost of standing still — falling behind competitors who modernized, paying for human-shaped software in an agent-shaped world — starts to exceed the cost of switching.

That she is talking about financial institutions matters. Banks and insurers are the most conservative software buyers on earth — the last to adopt, the most regulated, the most allergic to migration risk. If the calculus is flipping even there, it is flipping everywhere. Strange's framing is not that the new software is merely better; it is that not adopting it becomes the riskier choice. Once that perception takes hold among the most cautious buyers, the protective moat of inertia — the single biggest thing keeping incumbents safe — begins to drain.

Why Incumbents Are Structurally Vulnerable

The vulnerability is not that incumbents lack AI talent or budget. Most have both. It is that their entire product and business model is built around a unit that AI is making obsolete: the human seat. Legacy SaaS was designed for humans to log in and do work, and it was priced per logged-in human. The interface, the workflow, the data model, and the revenue line all assume a person sits at the center of the process. When an agent can do the work, every one of those assumptions becomes a liability rather than an asset.

This is the seat-pricing trap, and it runs deep. An incumbent earning most of its revenue per seat has a P&L that punishes it for reducing the number of seats its customers need — which is exactly what effective AI does. Deploy agents that let a customer do the same work with fewer people, and you have cut your own revenue. The incentive is to make AI a mild assist that preserves the seat count, not a genuine automation that collapses it. That is the precise opposite of what an unencumbered challenger will build. We traced the mechanics of this unraveling in our analysis of the collapse of per-seat SaaS pricing, where even entrenched vendors are reporting seat declines as agents replace licensed users.

Why Bolt-On AI Loses to AI-Native Rebuild

The incumbent's instinctive response is to bolt AI onto the existing product: add a copilot in the sidebar, sprinkle "summarize" buttons, ship an assistant that lives inside the legacy workflow. This feels like progress and demos well. It also loses, because it accepts the legacy workflow as a given and merely decorates it. The legacy workflow was designed for a human to perform a long sequence of steps. Adding AI to each step makes the human slightly faster at a process that, rethought from scratch, should not require the human at all.

The AI-native challenger does not decorate the old workflow — it deletes it. It asks what the customer is actually trying to achieve and designs the shortest path from intent to outcome, with agents doing the work and the interface shrinking to the points where a human genuinely needs to decide. The result is not the old screen plus a chatbot; it is a fundamentally thinner product where most of the steps the incumbent's UI was built to support simply no longer exist. You cannot reach that design by adding features to the old one. You have to start over.

This rebuild-versus-decorate distinction is the heart of the YC "SaaS Challengers" thesis. The opportunity is not to build a slightly better version of an incumbent's product with AI features. It is to rebuild the category AI-first, with the economics and workflow that the incumbent cannot adopt without dismantling its own business. The challenger's advantage is not a better model — every player has access to similar models. It is the freedom to redesign the whole thing around what the model makes possible.

Which Categories Fall First

Not every SaaS category is equally exposed. The first to face credible AI-native challengers share a recognizable profile: they are document-heavy (the core work is reading and producing text and forms), workflow-heavy (the product encodes a long, repeatable human process), and seat-priced (the incumbent charges per human user). Where all three overlap, the challenger has the most room — agents can do the document-and-workflow work, and the seat-priced incumbent has the most revenue to lose by automating it.

In practice that points at categories like back-office finance and accounting software, legal and contract tooling, customer support platforms, compliance and risk systems, recruiting and HR workflow tools, and large swaths of vertical SaaS in regulated industries. These are products where a human currently logs in, reads a lot, applies rules, and produces an output — precisely the work agents do well. The categories that fall later are those built on genuine network effects, deep proprietary data, or high-touch human relationships, where the software is incidental to a moat that AI does not erode.

It is worth being clear about what protects an incumbent and what does not. Switching cost born of pure inertia is eroding fast, as Strange's prediction implies. Switching cost born of real network effects or irreplaceable data is far more durable. An incumbent should be honest about which kind of moat it actually has, because mistaking inertia for a moat is exactly how comfortable market leaders get caught.

What Challengers Do Differently

The AI-native challenger's playbook is consistent across categories, and it is defined by three moves the incumbent struggles to copy. The first is outcome pricing: instead of charging per seat, the challenger charges per unit of delivered value — per contract reviewed, per ticket resolved, per filing produced. This is legible to buyers, aligns the vendor's incentives with the customer's, and sidesteps the seat logic entirely. The incumbent cannot easily follow without admitting its product needs fewer seats.

The second is agent-first workflow design: the product is built so that agents perform the work and humans supervise, rather than humans performing the work and agents assisting. The third is a thinner interface: because the agent does the steps, the UI does not need to expose every field and button the old human workflow required. The product becomes simpler to use precisely because it does more of the work itself. These three together produce a product that is cheaper to run, cheaper to buy on an outcome basis, and faster to deliver value — a combination the seat-priced incumbent cannot match without rebuilding.

The most aggressive challengers push this even further, past replacing the software and into replacing the service the software supported — doing the work outright and pricing on the result. That is the adjacent wave we examined in our piece on AI-native service firms that do the work rather than assist. A SaaS challenger and an AI-native service firm sit on a spectrum: both anchor value to the outcome, and both threaten incumbents whose model depends on humans in the loop being billed by the seat or the hour.

How Incumbents Should Actually Respond

The incumbent response that fails is the obvious one: ship a copilot, issue a press release, and hope the moat holds. That defends the seat revenue for a few quarters and loses the category over a few years. The response that has a chance is uncomfortable, because it requires attacking the very revenue model that made the incumbent successful. The honest move is to build the AI-native replacement for your own product before a challenger does, and to be willing to let it cannibalize your seat revenue — on the logic that it is far better to cannibalize yourself than to be cannibalized.

Concretely, that means standing up a genuinely independent AI-native product line, insulated from the pressure to protect the existing P&L, with permission to price on outcomes and rebuild the workflow from scratch. It means being clear-eyed about which of your moats are real network effects and data advantages versus mere switching inertia, and investing behind the former while assuming the latter will erode. And it means moving before the contract-renewal calculus flips for your customers the way Strange predicts it will for financial institutions — because once buyers decide that staying is the riskier choice, the incumbent advantage is already gone.

Conclusion: Inertia Is No Longer a Moat

The "SaaS Challengers" category exists because the oldest assumption in enterprise software — that incumbents are too entrenched to displace — is being tested by a technology that makes the incumbent's own structure a liability. Seat-based pricing, human-shaped workflows, and feature-laden interfaces were assets in the SaaS era and are anchors in the AI-native one. When the most conservative buyers on earth start treating modernization as the safer bet, as Strange predicts, the protective power of inertia is already failing.

None of this guarantees that every incumbent falls or every challenger wins; real moats built on data and network effects will hold, and many challengers will fail on execution. But the strategic terrain has shifted. The question for any software company in 2026 is no longer whether to add AI. It is whether you are willing to rebuild your product — and your pricing — around what AI makes possible, before someone with nothing to protect does it for you.