Agentic AI and the End of SaaS as We Know It

The world of B2B SaaS is on the cusp of a radical transformation. A new breed of agentic AI – AI that can reason, act autonomously, and deliver outcomes – is challenging the core assumptions of today’s software business models. Investors are starting to ask uncomfortable questions: What if instead of selling software seats, companies sell outcomes delivered by AI? What if autonomous software agents replace entire workflows that used to require dozens of SaaS apps and human users? In this thought-provoking analysis, we’ll explore how agentic AI and advanced reasoning models are redefining software consumption and value delivery, and why the next 2-3 years could upend the SaaS model that has dominated for two decades.

Expect some controversy – these ideas challenge the very foundation of recurring revenue SaaS. But as history shows, every tech paradigm meets its disruptor. For investors, understanding this shift is key to spotting the next winners (and avoiding the laggards) in enterprise software.

The Rise of Agentic AI: Software That Does the Work

Traditional software (even “intelligent” SaaS with basic automation) has always been a tool for humans to use. In contrast, agentic AI flips the script – it’s software that can do tasks autonomously. Instead of just providing an interface or analytics for a human, an AI agent can execute end-to-end workflows, make decisions, and continually learn from results. In effect, these AI systems function like digital employees or assistants, not just static tools.

This is made possible by new reasoning models (think GPT-4o and beyond) that handle complex sequences of actions. They can integrate with multiple software APIs, interpret data from different silos, and take actions to achieve a goal – all with minimal human input. Agentic AI doesn’t just assist work; it performs the work. As one industry analysis put it, this makes enterprise software more outcome-driven and autonomous, not merely a helper to humans. Early examples include AI agents that can onboard a new employee across HR, IT, and payroll systems automatically, or an AI that handles customer support tickets from start to finish.

This shift profoundly changes how software is consumed. If an AI agent can traverse many apps to complete a task, the user (the business) cares about the outcome – e.g. “employee onboarded” or “ticket resolved” – not which software tools were used along the way. In other words, the unit of value in software moves from usage (or seats) to results. It’s no wonder Nvidia’s CEO Jensen Huang riffed on Marc Andreessen’s famous line by saying “Software ate the world, and now AI is eating software,” signaling that AI is reshaping how software itself is built and delivered.

Cracks in the SaaS Model (and Why AI Exploits Them)

SaaS has been a fantastic business model over the last 15+ years – high margins, sticky subscriptions, and pricing power through per-seat licensing. But it also introduced new inefficiencies that businesses increasingly chafe at. Shelfware (unused subscriptions) and proliferation of apps have become major pain points. ?A 2022 survey found 42% of IT pros struggled to find and eliminate underutilized SaaS, with roughly one-third estimating 20–39% of SaaS spend was wasted on unused licenses. ?In tough economic times, CFOs don’t love paying for software people aren’t using.

Agentic AI strikes at the heart of this issue. If an AI can handle a task that might replace the work of, say, five people using five different apps, do you really need five separate seat licenses? Probably not. Fewer human operators means fewer seats. We’re already seeing early evidence of this “seat compression” in real organizations. For example, contact center software provider NICE reported a client was able to cut the number of agent licenses from 1,000 to 750 after rolling out NICE’s AI solutions – a 25% reduction in seats needed. Similarly, Salesforce’s own AI (Einstein) made some enterprise customers’ support teams ~10% more efficient, allowing them to reduce headcount (and Salesforce licenses) by 10% in those departments. In both cases, the AI took over enough work that fewer human users needed to log in – a direct hit on the seat-based SaaS revenue model.

Another crack in the model is the data silo and integration problem. Companies might use hundreds of SaaS apps, each a specialist, which don’t always talk to each other. Employees waste time stitching together outputs – one study showed knowledge workers spend 12 hours a week just “chasing data” across systems. Agentic AI has an answer here too: an autonomous agent can sit on top of all these apps and act as an orchestrator, pulling info from one system and inputting into another to complete workflows seamlessly. Instead of a human coordinator logging into 5 different tools, an AI agent with the right permissions can do it in seconds. This again shifts value away from the individual apps themselves and toward the agent that glues them together.

Bottom line: The traditional SaaS model – a myriad of separate apps, each charging per-user fees – is starting to strain under its own weight. AI is exposing the inefficiency of paying for software access rather than actual results. As one SaaS investor observed, selling software to let knowledge workers click around screens should eventually be massively disrupted by AI agents that just get the job done. If that sounds controversial, consider that it’s exactly what CEOs, CFOs, and customers increasingly want: don’t sell me a hammer, just nail the damn nail in the wall for me.

AI-Driven Shifts Already Happening in B2B SaaS

This isn’t in the future – we’re already seeing the first waves of AI-driven change in SaaS today. Savvy companies are experimenting with new offerings and pricing that reflect AI’s autonomous capabilities. Some notable examples:

• Customer Support on Autopilot: Customer support SaaS is ground zero for AI disruption. Intercom, for instance, introduced an AI chatbot agent called Fin that can fully resolve customer queries without human agents. How do they charge for it? Not by “seats” or monthly user fees, but per successful ticket resolution – roughly $1-2 per ticket solved by the AI. In other words, the customer only pays when the AI actually delivers the outcome (answering the question or fixing the issue). This is a prime example of outcome-based pricing emerging inside a traditional SaaS domain.
• Sales and Marketing Agents: A host of startups are launching AI-first services to handle sales outreach, lead generation, and other go-to-market tasks. For example, a startup called 11x.ai offers an AI Sales Development Representative named “Alice” that you effectively hire as a service. Instead of paying an annual SaaS fee, companies pay for an “AI SDR” unit that performs a defined amount of work – researching accounts, sending emails, booking meetings – analogous to a human SDR’s output. This blurs the line between software and labor; this is software behaving like a contract worker, with pricing to match the work done.
• HR and Employee Workflows: Even internal processes like HR are seeing AI agents step in. IBM has been developing Watson Orchestrate, an AI agent that can take on tasks like employee onboarding or scheduling meetings across various internal systems. And consider Workhuman, an HR SaaS platform: they moved to an outcome-based approach where they guarantee ROI on employee engagement metrics and don’t charge traditional seat fees. That essentially puts the onus on the software (often AI-driven) to deliver real improvements – if it doesn’t, the customer doesn’t pay full freight. This is a bold departure from the old “pay upfront and hope you get value” approach.
• Major SaaS Vendors Embracing AI Agents: Perhaps the biggest sign of change: incumbent SaaS giants themselves are preparing for an AI-agent future. In late 2024, Zendesk (a leader in customer service software) announced it will shift to outcomes-based pricing for its AI bots, saying it will soon “only charge for successful outcomes” achieved by its AI, rather than per-user or per-ticket fees. They openly acknowledged that traditional human-centric pricing doesn’t make sense when an autonomous agent is doing the work. Likewise, Microsoft and Salesforce have hinted that AI agents will become the backbone of future operations, fundamentally changing how they package value. It’s hard to overstate this: leading SaaS companies are effectively admitting that the old pricing models will break in an AI-driven world.
• All-in-One AI Platforms: New platforms are emerging that use AI to span across applications. OpenAI’s own ChatGPT, for example, now has a plugin with Zapier that lets it connect to 5,000+ apps and perform actions in them via natural language. This means a single AI interface can initiate workflows that touch your CRM, email, spreadsheets, project tools, you name it. When one AI agent can replace the need for a user to manually use dozens of different SaaS interfaces, it again undercuts the rationale for paying those individual subscription fees. Startups like Adept AI went further, building agents that literally operate software UIs like a human would (clicking buttons across multiple apps to get a job done) – a sign of how AI might fully abstract away app-by-app interaction. In fact, Adept’s vision was so compelling that in 2024 Amazon hired most of the Adept team to bolster its own AI capabilities. Big tech sees the writing on the wall.

In short, the shift has begun. We see AI reducing reliance on human software users, SaaS vendors piloting pay-for-results models, and investment (and acquisitions) pouring into AI-first solutions. These are the early indicators of a broader transformation.

New Business Models Poised to Emerge

What might the next generation of B2B software business models look like as this trend accelerates? Over the next 2–3 years, expect to see the rise of models that would sound radical by yesterday’s standards. Here are a few one might foresee:

1. AI-First Platforms as the New SaaS

Rather than software with an AI add-on, we’ll see platforms built from the ground up with an AI agent at the core. These “AI-first” platforms function more like an employee or a service provider than a piece of software. They might charge a monthly fee per AI agent you employ (akin to a salary but far lower cost than a human). For example, a company might subscribe to an “AI finance analyst” service that continuously analyzes data and generates reports, instead of buying a BI software license and hiring an analyst. The value proposition is directly achieving a business function out-of-the-box, not providing a toolset. This flips SaaS from product to “labor-as-a-service.” The closest current analogues are RPA (robotic process automation) bots and virtual assistants, but next-gen AI agents will be far more general and intelligent. Companies like Adept, Cognition Labs (with its AI engineer “Devin”), and others are pioneering here, effectively offering agents that can be customized for different roles.

Importantly, these platforms tend to be horizontal in nature – a single AI can handle multiple different tasks or workflows by spanning many apps. In the same way a smart human employee might wear multiple hats, an AI platform can be the ultimate generalist. This raises competitive stakes: why buy 5 different SaaS tools if one AI platform can do the jobs of all five? That’s exactly what agentic AI startups are betting on, and it’s a threat that incumbent vendors will need to navigate (likely by expanding their own AI capabilities or forming ecosystems so their apps play nice with AI orchestrators).

2. Outcome-Based (Pay-per-Result) Pricing

As hinted earlier, outcome-based pricing is about charging for actual value delivered rather than access or usage. While not entirely new (some services have done value-based pricing before), AI makes it far more practical. When an AI agent’s role is clearly tied to a business outcome – say, closing a deal, resolving a support case, or onboarding an employee – the vendor can price on that outcome. We’re already seeing toeholds: Intercom’s Fin at ~$2 per successful support answer, or Zendesk planning to charge only for issues actually solved by its bots. This model is highly appealing to customers, as it de-risks their spend (“we pay only if it works”). In a sense, it turns software purchasing into a variable cost aligned with success, not a fixed cost.

For SaaS vendors, of course, this is a huge shift – it transfers some risk to the vendor and may make revenue less predictable. But it can also vastly expand adoption (who wouldn’t try a service that you only pay when it succeeds?). Vendors that nail outcome-based pricing will foster deeper trust with customers, almost operating as partners. In fact, industry consultants note that in the agentic AI era, successful software firms will evolve into partners focused on outcomes rather than mere software providers (see my colleague Barak Ravid’s perspectives on this trend here.)

Analogies from other industries abound here. Think of how digital advertising moved from paying per ad impression (old model, like paying per user login) to paying per click or per action (new model, like paying per conversion). Advertisers love paying per action (CPA) because it guarantees an outcome, whereas publishers prefer the old CPM model because it guarantees their revenue whether or not the ad works. In software, traditional SaaS is like the publisher charging CPM – “pay me even if you don’t fully use the product” – whereas outcome-based AI services are like CPA – “pay us only if we deliver your desired result.” It’s easy to see which way the wind is blowing when buyers gain bargaining power. A recent report even noted that Zendesk and others plan to “eschew seat-license fees” in favor of charging only for successful outcomes, believing customers will soon demand this alignment of cost to value.

It should be noted, outcome pricing isn’t trivial to implement. Vendors and customers will have to agree on metrics (e.g. what counts as a resolved support ticket, or how to measure a sales lead converted to a deal) and handle attribution fairly. There’s risk of disagreements if the outcome has many contributing factors. To mitigate this, some companies might start with usage-based proxies (e.g. charging per AI task performed, which is easier to count) as a stepping stone. But over time, as trust builds and technology for tracking outcomes improves, more contracts will move to true pay-for-performance. Investors should watch for SaaS sales models starting to resemble consulting or outsourcing contracts with SLAs and outcome guarantees – that’s a clear signal of this trend.

3. Fully Autonomous Workflows (No Human in the Loop)

Today, even when AI is used, there’s often a human overseeing or providing input (the classic “human in the loop”). The next leap is software that operates with full autonomy for defined tasks. In certain domains, we’re close to that. For instance, some finance software can already run autonomously to detect and reconcile billing errors, only alerting a human if something truly novel occurs. As agentic AI matures, we’ll see AI-run departments in a company – perhaps a fully autonomous accounts payable process, or an AI-managed IT support system that handles routine issues end-to-end and orders its own replacements.

When software runs with minimal human oversight, the business model might resemble a utility or an outsourced service. A company might pay a monthly service fee for an “autonomous IT helpdesk” agent that resolves, say, 1,000 tickets a month on its own. If more tickets come in, the AI auto-scales and the bill goes up (similar to cloud computing usage scaling). If volume is low, the company pays less. This is essentially utilities-style consumption applied to business processes – a direct analog to cloud platforms charging by compute/hour or storage/GB, but now it could be “tickets resolved” or “invoices processed” as the meter.

One fascinating example: Hippocratic AI is developing autonomous healthcare agents to handle certain non-diagnostic tasks that nurses normally do. They claim their AI “virtual nurse” can perform specific routine tasks at an equivalent cost of about $9/hour vs. $60/hour for a human RN. They’re so confident in the AI’s autonomy and reliability that they’re essentially pricing it like labor (and at a massive discount). While healthcare is a specialized case, it showcases how a fully autonomous agent can completely change cost structures. In enterprise SaaS, any workflow that can be codified and learned by AI is a candidate for this kind of disruption.

Analogies from Past Disruptions: Clues to the Future

If all this sounds unprecedented, remember we’ve seen business model disruption in tech before. History doesn’t repeat, but it often rhymes:

• On-Prem to Cloud: Two decades ago, traditional software was sold via hefty up-front licenses and annual maintenance – until Salesforce and the SaaS wave turned it into a subscription. Many incumbent software companies struggled with this shift (the Innovator’s Dilemma in action), while new entrants seized the moment. The result: a massive value migration from old guard to cloud-native firms. Today, AI-first upstarts are in a similar position relative to SaaS incumbents. Those stuck purely on seat-based subscription models might go the way of legacy license software providers if they can’t adapt.
• Cloud to Serverless/Utility: Even within cloud computing, we went from renting virtual machines (you manage the servers) to serverless functions (just run my code, abstract everything else). Each step moved the provider-tenant relationship to a higher-level outcome. We can view AI agents as the “serverless” model of software usage – you give the AI a goal, it figures out all the steps (calls various apps, databases, etc.) to deliver that result. Just as cloud utilities changed cost models (from fixed server instances to pay-per-use), AI agents could turn SaaS into pay-per-business-transaction.
• Advertising (as noted earlier): The shift from paying for exposure to paying for success in ads foreshadows the pressure software vendors will feel. Customers will increasingly ask, “Why should we pay for 100 users on a system if we only see 70 units of value? Let’s structure the deal so we pay for those 70 units of actual value.” It’s a buyer’s market mentality that puts incumbent pricing models under stress, much as ad buyers pushed for CPA deals over CPM.
• Automation in Manufacturing: A non-software analogy: factories once paid armies of workers; then robots and machines took over repetitive tasks. Today, factories effectively pay for throughput – if a machine can output 100 units an hour, they care about that output, not how many workers it would have taken. In the white-collar world, AI agents are the new robots, and businesses will start measuring their software costs in terms of work throughput (like tickets resolved, reports generated) rather than users logged in. The companies that provided manual labor (outsourcing firms, etc.) had to reinvent themselves for the automation age or lose out – similarly, SaaS firms must reinvent around AI or risk irrelevance.
• The Demise of Once-Dominant Players: We’ve seen entire sectors of tech fall when paradigms shift – consider how many once-dominant networking hardware companies got decimated when the industry moved towards software-defined networking and cloud services. Some investors are openly asking if the 100+ public SaaS companies today could face a similar fate, ending a 20-year golden era. It’s an extreme scenario, but not impossible if they fail to adapt. Of course, incumbents aren’t standing still (we discussed how some are integrating AI or changing pricing), so the more likely outcome is a mix of evolution and creative destruction.

Timeline: How Soon Will Disruption Hit (and What Investors Should Watch)

Is this all going to play out overnight? Unlikely – enterprise software rarely changes in a flash. However, the pace of AI adoption is far faster than past tech shifts, and we’re already in the early innings of change. The next 2-3 years will be critical. Here’s a forecast and key signals investors should monitor:

• Major SaaS Vendors Adapting (2024-2025): We’re already seeing first movers like Zendesk and Workhuman with outcome-based models. If in the next 12-18 months more big SaaS names (Salesforce, ServiceNow, Microsoft, etc.) announce AI-driven pricing or product changes, that’s a strong signal. For instance, if Salesforce were to introduce a plan where you pay per AI-completed service case (instead of per agent license), it would validate that the model is shifting. Keep an eye on earnings calls and product launches for language around “usage-based” or “outcome-based” offerings.
• Customer Behavior – Seat Saturation: Investors should watch metrics like net seat expansion and upsell in traditional SaaS. If enterprises start trimming licenses because AI features let them do more with fewer people, it will appear as slower seat growth or even downsizing in SaaS deployments. The anecdotes of 10-25% seat reductions could become more widespread, pressuring SaaS vendors to find new revenue streams (like charging for AI or usage). If you hear about a Fortune 500 firm renegotiating a big SaaS contract to cut users due to AI efficiencies, that’s a harbinger.
• AI Native Startup Traction: The market is currently flooded with AI startups, but only some will break out. Investors should look for AI-first SaaS challengers winning big enterprise deals or raising significant rounds at high valuations. Notable examples to date: Hippocratic AI raised $500M for its AI health agents, indicating huge confidence; Cognition Labs at a $2B valuation for AI software engineers; and the aforementioned Adept being snapped up by Amazon. If an AI startup begins stealing customers from an established SaaS player by offering a usage or outcome-based model (e.g. an AI sales agent platform vs. a traditional CRM module), that incumbent is in trouble. Early partnerships between incumbents and AI startups can also signal who’s aligning with the trend instead of fighting it.
• Pricing Innovation and Experiments: Keep track of how software pricing models evolve. More adoption of consumption pricing in SaaS contracts (beyond just cloud infrastructure usage) will show that customers demand flexibility. For example, if a vendor like Atlassian or Oracle starts offering an “outcome guarantee” or metered pricing based on tasks completed, it would have been unthinkable a few years ago – a real sign of the times. Any move away from pure per-seat subscriptions is noteworthy. Analysts are already noting that many SaaS firms are quietly adding hybrid pricing (part seat, part usage) to not be left behind.
• AI Performance Benchmarks: The technical side matters too. Watch for AI agents hitting new levels of reliability on complex tasks. The closer they get to human-level competence (or surpass it) in various workflows, the faster companies will trust them with autonomy. For instance, if an AI can handle, say, 90% of Tier-1 support tickets correctly, that might be the tipping point where a company feels safe to reduce its support staff and rely on the AI. Successful case studies (with hard ROI numbers) of AI agents in action will catalyze faster business model change. Conversely, if high-profile AI failures occur, it might slow adoption in the short term (though likely just a speed bump). As an investor, one might even develop an “AI Disruption Index” of sorts – tracking enterprise AI capability milestones that correlate with SaaS usage changes.

Opportunities and Risks for Investors

For investors, this coming shift presents both huge opportunities and significant risks:

• Opportunities: The obvious winners are the companies building and embracing these new models. AI-first SaaS startups that crack the code on outcome-based offerings could scale at the expense of incumbents. Even incumbents that successfully reinvent themselves (e.g. by launching their own agentic AI platforms or restructuring pricing) could tap into massive new addressable markets. ?If software can truly deliver more direct value, enterprises might actually spend more on those solutions (imagine software that directly drives revenue or cuts major costs – there’s immense willingness to pay for that). Investors should look for businesses that position themselves as partners in their customers’ success – those will earn loyalty and premium economics. Additionally, the infrastructure and tooling around agentic AI (from AI model providers to integration platforms and security for AI agents) will be a fertile ground, analogous to how picks-and-shovels plays thrived during the SaaS and cloud booms.
• Risks: On the flip side, many traditional SaaS players could see their moats erode. If your value rests on a nice UI and lots of human users, an AI that bypasses the UI and handles tasks could make you irrelevant. Revenue models built on per-seat pricing face a reckoning – companies that resist change may experience slowing growth or customer churn as more agile competitors offer outcome-based deals. There’s also execution risk: transitioning a business model (say from subscription to usage pricing) is hard on a company’s financials and sales organization. It might depress short-term revenue or require new ways to market and sell. Not every company will navigate that gracefully (imagine the first major SaaS firm to announce a shift from stable annual subscriptions to pay-per-use – Wall Street might panic in the short term). Investors need to discern which companies have the DNA to innovate on business model, not just add a sprinkle of AI to their product. A good heuristic is to listen for management teams talking about value delivered and customer ROI, rather than just touting AI features. If a SaaS CEO is still fixated on selling more seats while a competitor is selling outcomes, that’s a red flag.

Conclusion: A New Chapter for Software – Adapt or Be Disrupted

The age of agentic AI is poised to rewrite the rules of enterprise software. While it’s tempting to dismiss some of this as hype (after all, core IT systems don’t get ripped out overnight), the strategic direction is becoming clear. Software that can autonomously deliver business outcomes changes the value equation for customers. B2B buyers will favor vendors who take on more risk and accountability for results. This will reward a new generation of AI-driven platforms and could punish those SaaS companies that cling to legacy models of selling seats and features.

Investors should prepare for some volatility as this unfolds – we may see unusual pricing schemes, shorter initial contracts, or lower traditional SaaS metrics in the transition. But we may also see the emergence of a few dominant AI-first “operating systems” for business that command platform-like multiples. As with any disruption, timing is tricky: too early, and you invest in science projects that don’t commercialize; too late, and you’re holding the next Blockbuster Video of software. The 2-3 year horizon will reveal a lot about who’s adapting.

One thing is certain: the conversation in boardrooms has shifted from “How can we add AI to our product to charge more?” to “How do we rethink our entire business around AI delivering value?” ?The SaaS revolution ate the world of on-prem software; now AI is coming to eat SaaS. For those investors and founders that recognize what’s happening, it’s an exciting (if slightly unnerving) time – a chance to back the next winners who turn this agentic AI disruption into decades of growth, and to avoid those stuck fighting the last war. In the end, companies that embrace delivering real outcomes – not just software access – will build deeper customer relationships and unlock greater long-term value. ?The rest may face an existential challenge. Controversial as it sounds, in a few years we might look back and realize that the service in “Software-as-a-Service” was always meant to be taken literally – service, not software, is what the customer really wanted all along.