AI Is Eating Software: Is B2B SaaS Dead?

For most of the last two decades, B2B SaaS was one of the most reliable business models in technology.

You sold a seat. You renewed the seat. You expanded the account. You raised price modestly every year. If you had strong retention, low churn, and credible net revenue expansion, the market rewarded you with premium multiples. Public investors loved the predictability. Private investors loved the repeatability. Founders loved the playbook. And over time, an entire generation of enterprise software companies was built around a simple assumption: great software meant predictable revenue per user.

That assumption is now under pressure. Not because software is going away. It is not, but the unit of value is changing. The old model sold access to tools and the new model increasingly sells completed work.

That distinction sounds subtle, but it isn’t. It is the difference between paying for a CRM seat and paying for qualified pipeline created. Between paying for a support platform and paying for resolved tickets. Between paying for an analytics dashboard and paying for the answer. Between buying software that helps a human do work and buying an AI system that performs part of the work directly.

That is the real reason the market keeps asking whether B2B SaaS is dead.

My answer as an investor is no. But the version of SaaS that minted Salesforce, Workday, ServiceNow, Atlassian, and dozens of other category-defining companies is being rewritten in real time. Capital is reallocating. Pricing models are changing. Distribution is shifting. Buyers are rethinking what they are willing to pay for. And a new generation of AI-native companies is scaling faster than anything the software category has produced before.

Investors who understand what is actually changing and what is not will be positioned for the most interesting decade in enterprise software since the original cloud transition.

The headline is simple: SaaS is not dying. The middle is getting squeezed.

What The Data is Telling Us

The numbers from the past 18 months read less like a normal cycle and more like a regime change.

On the funding side, AI has become the center of gravity for venture capital. AI captured roughly half of all global venture funding in 2025 and around 80% of it in Q1 2026, driven by mega-financings for OpenAI, Anthropic, and xAI (Crunchbase News, 2026). Agentic AI, the category most directly competitive with traditional seat-based SaaS, raised $6.42 billion in 2025, the largest year ever for autonomous software (AgentMarketCap, 2026). Gartner now projects worldwide AI spending will reach $2.5 trillion in 2026 (Gartner, 2026), and IDC expects agentic AI to capture more than a quarter of global IT spend, roughly $1.3 trillion, by 2029 (IDC, 2025).

Those are not normal adoption curves. They are reallocation curves. The public markets are sending the same message from the other direction. The median enterprise-value-to-next-twelve-months-revenue multiple for cloud software fell to 3.6x in early 2026, the lowest level in a decade and down from a 2021 peak above 18x. For the first time in history, software’s forward P/E dropped below the S&P 500, falling from 84x at the 2020–2022 high to roughly 23x by March 2026 (Jamin Ball, Clouded Judgement, 2026). The early-2026 software selloff, what Marc Benioff dryly called “not my first SaaSpocalypse,” wiped out roughly $285 billion of software-stock value in a matter of weeks (Salesforce Ben, 2026).

That is not just a valuation reset. It is the market asking a more fundamental question: if AI can perform work that software previously only helped coordinate, what should software companies be worth?

The market is asking the question: if AI can perform work that software previously only helped coordinate, what should software companies be worth?

Then there are the AI-native growth curves. They are almost absurd by historical SaaS standards. Cursor went from $100 million ARR in January 2025 to $2 billion in February 2026 (TheNextWeb, 2026). Anthropic grew enterprise revenue 80x in 15 months to a $30 billion run-rate by April 2026 (VentureBeat, 2026). Sierra reached $100 million ARR in under two years from founding (TechCrunch, 2025). Glean doubled from $100 million to $200 million ARR in nine months (Fortune, 2025). Harvey reached $190 million ARR by January 2026 and raised at an $11 billion valuation in March (Harvey, 2026). ElevenLabs crossed $500 million ARR in early 2026 (CNBC, 2026).

In the prior SaaS era, getting to $100 million ARR in five to seven years was elite. Now we are seeing AI-native companies hit that milestone in two years, sometimes less, and continue compounding at a pace that breaks most historical venture benchmarks.

Some of this is hype. Some of it is pull-forward. Some of it will disappoint. But dismissing the entire pattern as a bubble is lazy. The more useful interpretation is that enterprise buyers are not merely buying new software. They are buying a new labor architecture.

That is a much larger market.

Why The Question is Being Asked Now

The reason the “Is SaaS dead?” question has become unavoidable is that some of the smartest operators in the industry have started saying the quiet part out loud.

Satya Nadella said in late 2024 that “the notion that business applications exist” may collapse in the agent era (Windows Central, 2024). His point was not that databases disappear. His point was that many business applications are essentially CRUD (create, read, update, delete) databases wrapped in workflow logic. If the logic moves to agents, the application layer changes. The agent becomes the interface. The system of record becomes infrastructure. The old app becomes less central.

Bret Taylor, who built Sierra after running Salesforce, has made a similar argument from a different angle: the atomic unit of AI is the agent, and the vast majority of digital interactions will eventually happen through agents (Stratechery, 2025).

Martin Casado at a16z has said that venture capital itself is being legitimately disrupted as a discipline. Benedict Evans has gone even further in his read on the Series A market, arguing that it has become virtually impossible to raise a Series A for a traditional SaaS company without an AI-native thesis (Business Insider, 2025).

When the CEO of Microsoft, the founder of Sierra, a top-tier venture investor, and one of the sharpest technology analysts in the world are all describing the same shift, the right question is not whether change is coming.

The right question is what survives it.

A Framework: What Gets Eaten, What Survives, What Gets Bigger

Strip away the noise and three patterns explain most of what is happening in enterprise software right now. Some categories get eaten. Some survive. Some get bigger.

The mistake is treating all SaaS as one thing. It never was. “B2B SaaS” includes lightweight horizontal productivity tools, regulated systems of record, vertical operating systems, workflow automation, data infrastructure, developer tools, security platforms, finance systems, and industry-specific applications. AI will not affect all of those categories the same way.

The right analysis is category by category, workflow by workflow, and pricing model by pricing model.

1. What Gets Eaten:

The most vulnerable software is horizontal, undifferentiated, seat-based tooling whose value was always something like: we put a better UI on a database and made a human workflow easier to manage.

That was enough in the cloud era. It is not enough in the agent era.

The classic example is customer support. For years, support software sold seats to agents and workflows to managers. The more humans in the loop, the more software you could sell. AI flips that logic. Intercom now sells Fin at $0.99 per resolved outcome, with average resolution rates above 67%. Sierra prices around $1.50 per resolution and only charges when its agent solves the issue end-to-end.

That is a different economic proposition. The buyer is not paying for access. The buyer is paying for work completed.

Once that happens, every horizontal SaaS company priced like a labor tax has a problem. If I can pay per resolution, why should I pay per support seat? If I can pay per qualified lead, why should I pay per sales-development seat? If I can pay per document reviewed, why should I pay for a research seat? If I can pay per workflow completed, why should I pay for a human-facing dashboard that merely coordinates the workflow?

This does not mean all seats disappear. But it does mean the seat is no longer the default unit of value.

The pricing data already shows the shift. Seat-based pricing dropped from 21% to 15% of B2B software companies in just twelve months, while hybrid models surged from 27% to 41% (Bain, 2025). Gartner expects more than 40% of enterprise SaaS spend to move off pure seat-based pricing by 2030.

That is the leading indicator. Pricing changes before company categories change. When buyers stop wanting to pay for seats, the market is telling us what it values instead.

It values outcomes.

This is why the undifferentiated middle of SaaS is so exposed. If a product is horizontal, lightly integrated, weakly differentiated, and primarily sold on seat count, AI compresses it from both sides. Foundation models commoditize the intelligence layer. Agents absorb the workflow layer. Incumbents bundle similar functionality into broader platforms. And buyers start asking why they are paying for another login.

That is a bad place to be.

2. What Survives:

The second category is software that survives and in many cases becomes more valuable.

Workflow-deep, vertical, regulated, system-of-record software still has real moats. In some categories, those moats are strengthening.

There are a few reasons.

First, compliance matters. HIPAA, Basel III, FedRAMP, FINRA, SOC 2, state-level privacy regimes, procurement rules, auditability, legal accountability — these are not small obstacles. The closer software sits to regulated decisions, financial workflows, legal liability, or operational risk, the harder it is to replace with a probabilistic agent.

Second, systems of record are sticky for a reason. They hold canonical data. They connect to other systems. They encode permissions. They create audit trails. They become part of how an organization operates. An AI agent may become the interface, but the underlying system of record still matters. In fact, it may matter more, because agents are only as useful as the data, permissions, and workflow context they can access.

Third, proprietary data becomes more valuable when models commoditize. If everyone has access to similar foundation models, differentiation moves to data, workflow context, integrations, and trust. A vertical software company with years of proprietary operational data in construction, healthcare, logistics, insurance, payments, or legal workflows has an asset that a generic horizontal AI wrapper does not.

This is why the “AI replaces SaaS” narrative is too crude. In many enterprise contexts, AI does not replace the system of record. It layers on top of it. The agent becomes the interaction layer. The workflow may become more automated. The human interface may change dramatically. But the data model, permission structure, compliance layer, and audit trail remain essential.

That is the durable middle of the software stack.

Not all vertical software wins, of course. Bad vertical software will still get punished. Thin vertical SaaS with weak data, shallow workflows, and limited customer love is vulnerable. But deeply embedded vertical operating systems are in a much better position than generic horizontal tools.

3. What Gets Bigger:

The third category is the one many people are underestimating: incumbents with distribution and data can get bigger.

The lazy version of the AI disruption story says incumbents lose and startups win. Sometimes that happens. But in enterprise software, distribution matters enormously. Existing customer relationships matter. Installed bases matter. Data gravity matters. Procurement trust matters. Security approvals matter. Integration surfaces matter.

Salesforce’s Agentforce hit $800 million ARR, up 169% YoY, with 29,000 deals closed in a single quarter and 60%+ of bookings coming from existing customers (Salesforce IR, 2026). ServiceNow’s Now Assist crossed $600 million ACV and is on a $750 million run-rate entering 2026, with management raising the full-year AI projection to ~$1.5 billion (ServiceNow IR, 2026). Microsoft Copilot is now a $14 billion+ annualized business with M365 Copilot at 15 million paid seats, up 160% YoY (Microsoft IR, 2026). Adobe’s AI-influenced ARR exceeds $5 billion and Firefly is used by 75% of the Fortune 500 (Futurum, 2025).

These are not defensive numbers. They are offensive numbers.

The reason is straightforward. Large software incumbents sit inside existing enterprise workflows. They have distribution. They have procurement access. They have trust. They have data. They can bundle AI into existing contracts, upsell existing customers, and convert distribution into agentic revenue.

That does not mean every incumbent wins. Many will ship mediocre copilots, overcharge for them, and eventually face churn. But the best incumbents are not standing still. They are using AI to deepen their relevance, expand ACV, and defend the system-of-record layer.

The result is not a clean startup-versus-incumbent story. It is a barbell.

On one end, AI-native companies attack specific workflows with dramatically better products and outcome-based economics. On the other end, scaled incumbents use distribution and data to absorb AI demand across massive installed bases. The compressed zone is the undifferentiated middle: generic software without startup speed or incumbent distribution.

That is where capital is leaving fastest.

What We Are Actually Investing In

As an investor with the Alumni Ventures Seed Fund, we are building our AI portfolio around this exact thesis.

The new winners in enterprise software look less like traditional horizontal seat-based SaaS and more like agentic platforms, vertical operating systems, and infrastructure companies powering both.

A few examples from our Alumni Ventures portfolio illustrate the pattern.

Agentic Platforms Replacing Seats

Tabs is automating contract-to-cash for B2B revenue operations — the AR, billing, and collections workflows historically handled by a patchwork of point tools and human finance-ops teams. This is exactly the kind of back-office function where AI can move software from “helps humans manage work” to “performs meaningful portions of the work.”

Wispr AI is rebuilding the input layer of productivity software around voice and intent. That may sound narrow at first, but the input layer is one of the most important interfaces in computing. If humans interact with software less through clicking and typing and more through speech, intent, and context, then the old UI model becomes less central.

These are not AI features bolted onto SaaS. They are products designed under the assumption that agents, not seats, become the unit of value.

That matters. In the last era, founders often started with a workflow and asked, “How do we sell this as software?” In this era, the best founders start with a job to be done and ask, “How much of this work can the system complete?”

That is a different company-building motion.

Vertical AI in Regulated and Legacy Industries

Patlytics is applying AI to patent and IP intelligence, a market historically dominated by clunky, expensive legal-research software. This is a strong example of where vertical context matters. Legal workflows are not just generic document workflows. They require domain knowledge, source reliability, auditability, and trust.

Trunk Tools is bringing intelligent document and execution workflows to construction, a $13 trillion industry where much of the operational layer still runs through spreadsheets, PDFs, email, and tribal knowledge. Construction is not an easy market, which is exactly why the right vertical AI company can build a durable position. The workflows are messy. The data is fragmented. The stakes are high. Generic tools rarely go deep enough.

KarmaCheck is reshaping background checks and credentialing. Hayden AI is replacing legacy municipal enforcement software with autonomous vision. These are not generic AI wrappers. They are vertical systems built around data, workflow, compliance, and real-world operational constraints.

That is the key. Vertical context plus AI is harder to compete with than horizontal AI alone. The data, integrations, customer trust, regulatory posture, and workflow depth are co-located. That creates defensibility.

AI Infrastructure: The Picks and Shovels

The application layer gets the headlines, but infrastructure is where some of the most durable value may accrue.

Cohere builds enterprise LLMs with private, secure deployment — exactly the configuration regulated buyers need. Not every enterprise wants to send sensitive data to a general-purpose consumer AI model. Many need private deployment, security controls, and enterprise-grade reliability.

Groq’s LPU architecture delivers AI inference at a different cost curve than incumbent GPU clouds and now serves more than 650,000 API users. Inference cost is one of the central bottlenecks in AI adoption. If agents are going to run continuously across enterprise workflows, cost and latency matter enormously.

Lambda’s GPU cloud has become a multibillion-dollar partner to Microsoft and is widely expected to IPO. Together AI is building the open-source AI cloud — an AWS-like layer for models that enterprises increasingly want to run on their own terms.

These companies matter because the AI application boom requires a new infrastructure stack. Compute, inference, model hosting, data preparation, orchestration, observability, security, and deployment are all being rebuilt. The companies that power that layer can capture value regardless of which specific applications win.

In other words, when the software stack shifts, the picks and shovels matter.

The Data and Knowledge Layer

AI is only as good as the context it can access.

That makes the data and knowledge layer one of the most important parts of the enterprise AI stack.

Unstructured is becoming a default data-prep and ETL layer for enterprise RAG pipelines. This is not glamorous work, but it is essential. Enterprises have mountains of unstructured data sitting in PDFs, emails, documents, tickets, transcripts, contracts, manuals, and internal knowledge bases. Turning that into usable AI context is a massive problem.

Mercor went from roughly $1 million to $500 million of annualized run-rate in 17 months by supplying expert human data to frontier AI labs. That growth says something important: even in a world of increasingly powerful models, high-quality human expertise remains scarce and valuable.

You.com is building enterprise AI search and reasoning agents at a moment when internal search is being reimagined wholesale. Traditional enterprise search was about finding documents. AI-native search is about synthesizing answers, reasoning across sources, and taking action.

This layer will be critical because enterprises do not merely need models. They need models connected to their own knowledge, permissions, data, and workflows.

AI-native Creative and Content Tooling

Opus Clip is another useful example. Its generative-video editing tools are used to produce hundreds of millions of clips a year, eating directly into traditional creator SaaS workflows.

This is a pattern we should expect to see across creative software. The old product helped a human edit. The new product performs a meaningful part of the editing. That changes who can create, how fast they can create, and what they are willing to pay for.

Again, the question is not whether there is still software. Of course there is. The question is whether the software is selling access to tools or selling creative output.

The latter is a bigger market.

B2B Software Retooled, Not Replaced

Not every strong company needs to be “AI-native” from day one. Some of the most interesting opportunities are companies with strong existing positions that can use AI to deepen their value.

Enable’s rebate-management platform is launching an AI-native layer on top of category-leading distribution. Tilled is enabling vertical SaaS companies to embed payments — increasingly the economic moat that lets vertical software survive an AI margin squeeze. Cents is a vertical operating system for laundromats — effectively a Toast for laundry — that turned profitable in 2025 by being indispensable to operators rather than competing on per-seat economics.

These companies show why the “SaaS is dead” narrative misses the point. The durable companies are not necessarily the ones with the loudest AI branding. They are the ones that become indispensable to the workflow, own economic leverage points, and can use AI to increase value per customer.

The best B2B software companies will not be replaced by AI. They will be retooled around it.

Reasons To Be Cautious

A balanced view requires sitting with the parts of this story that are not yet settled.

The first caution is ROI. McKinsey’s 2025 State of AI survey found that 88% of organizations now use AI in at least one function, but only about 6% are seeing meaningful financial returns. Adoption is broad (McKinsey, 2025). Realized ROI is narrow.

That gap matters.

It tells us that many enterprises are experimenting, piloting, and deploying AI before they have fully redesigned workflows around it. That is normal in a platform shift. But it also means not every AI product with early revenue has durable value. Some tools will be tried and abandoned. Some copilots will not justify their price. Some agentic workflows will break in production. Some customers will discover that automation without process redesign does not deliver the promised economics.

The second caution is quality. Klarna’s high-profile experiment — replacing 700 customer service agents with AI and ending its Salesforce and Workday relationships — partially reversed in 2025 after quality issues. The company rehired human agents to staff a tiered support model.

The lesson is not that AI failed. The lesson is that “rip and replace” is the wrong mental model.

The better model is composition: agents on top of systems of record, humans at the edges, escalation paths where judgment matters, and continuous feedback loops that improve the system over time.

Enterprises do not just need automation. They need reliable automation. That is harder.

The third caution is valuation. Parts of this market are showing real bubble dynamics. Forward-revenue multiples of 50x to 100x for the fastest-growing AI-native companies assume execution paths that not every company will achieve. Some will. Many will not.

Diligence on durability matters more now than it has in any prior software cycle. Retention matters. Gross margin trajectory matters. Defensibility against foundation model improvements matters. Distribution matters. Data access matters. Workflow depth matters. The ability to move from pilot to production matters. The ability to price on outcomes without destroying margins matters.

In a market this hot, capital will fund many companies that should not exist. That is not a reason to avoid the category. It is a reason to underwrite carefully.

From a venture standpoint, this is exactly the environment where active selection matters. When a category is repricing, dispersion widens. The difference between the winners and the median company becomes enormous. That is when manager skill, access, and stage-appropriate underwriting matter most.

The old SaaS playbook was about selling access. The new enterprise software playbook is about delivering work.

What This Means for Investors

As an investor with the Alumni Ventures Seed Fund, I’d reduce this transition to three takeaways. First, B2B SaaS is not dying. It is bifurcating.

The horizontal, seat-priced, undifferentiated middle is being compressed. Verticalized, workflow-deep, system-of-record software remains valuable. And the agent-and-infrastructure layer is producing a generation of companies that scale faster than anything the category has historically produced.

The right question is not “Do we still invest in SaaS?” The right question is “Which kind of software company is this?”

Is it a thin workflow tool? Is it a system of record? Is it a vertical operating system? Is it an agentic labor replacement? Is it infrastructure? Does it own unique data? Does it have distribution? Does it produce an outcome buyers can measure? Does AI increase its value or commoditize it?

Those questions matter more than the label.

Second, pricing is the leading indicator.

Watch how a software company sells before you obsess over what it sells. The shift from seat-based pricing to usage, hybrid, agent, and outcome-based models tells you whether a company is positioned for the next decade or fighting the last one.

Seats are not going away entirely. But pure seat-based pricing is becoming harder to defend in categories where AI can complete work directly. The companies that make the transition well will align price with customer value. The ones that do not will face margin pressure, churn, and valuation compression.

By 2030, Gartner expects at least 40% of enterprise SaaS spend to be priced on usage, agent, or outcome models. I would not be surprised if the best AI-native companies get there faster.

Third, distribution and data still matter, sometimes more than ever.

GitHub Copilot’s share of AI coding assistants is not just about model quality. It is also about distribution through GitHub and VS Code. Salesforce’s Agentforce growth is overwhelmingly inside its existing base. Microsoft Copilot benefits from Microsoft’s position inside the enterprise. Adobe Firefly benefits from Adobe’s creative installed base.

The companies most likely to win the long arc of this transition are either AI-native builders with sharp vertical or workflow focus, or incumbents that successfully convert distribution into agentic revenue.

The middle – generic horizontal SaaS without either advantage – is where capital is leaving fastest.

The Bottom Line

The headlines will keep oscillating between “SaaS is dead” and “SaaS just had its best quarter.” Both can be true at the same time, depending on which version of SaaS we are talking about.

Traditional horizontal seat-based SaaS is under real pressure. Workflow-deep vertical software is still durable. Incumbents with distribution are not dead; the best of them are monetizing AI aggressively. AI-native companies are scaling at unprecedented rates. Infrastructure is being rebuilt. Pricing is moving toward usage and outcomes. Buyers are increasingly asking whether software can do the work, not just help humans manage it.

That is the shift.

The old SaaS playbook was about selling access.

The new enterprise software playbook is about delivering work.

For investors, that means the opportunity is not gone. It is moving. The next decade of value will accrue to companies that rebuild around AI from the ground up, own deep workflow context, capture proprietary data, align pricing with outcomes, and use distribution intelligently.

SaaS is not dead.

But the era of selling another seat to another dashboard and calling it a category may be ending.

The arsenal of enterprise software is being rebuilt. The companies that get rebuilt around AI — not patched with AI — are where the next decade of value is most likely to live.

 

This communication is from Alumni Ventures, a for-profit venture capital company that is not affiliated with or endorsed by any school. It is not personalized advice, and AV only provides advice to its client funds. This communication is neither an offer to sell, nor a solicitation of an offer to purchase, any security. Such offers are made only pursuant to the formal offering documents for the fund(s) concerned, and describe significant risks and other material information that should be carefully considered before investing. For additional information, please see here. Achievement of investment objectives, including any amount of investment return, cannot be guaranteed. Co-investors are shown for illustrative purposes only, do not reflect all organizations with which AV co-invests, and do not necessarily indicate future co-investors. Example portfolio companies shown are not available to future investors, except potentially in the case of follow-on investments. Venture capital investing involves substantial risk, including risk of loss of all capital invested. Diversification cannot prevent investment loss; it is a strategy to mitigate investment risk. This communication includes forward-looking statements, generally consisting of any statement pertaining to any issue other than historical fact, including without limitation predictions, financial projections, the anticipated results of the execution of any plan or strategy, the expectation or belief of the speaker, or other events or circumstances to exist in the future. Forward-looking statements are not representations of actual fact, depend on certain assumptions that may not be realized, and are not guaranteed to occur. Any forward-looking statements included in this communication speak only as of the date of the communication. AV and its affiliates disclaim any obligation to update, amend, or alter such forward-looking statements, whether due to subsequent events, new information, or otherwise.

Interested in Seeing Elite Venture Deals?