Will Agentic AI Disrupt SaaS?

A scenario-based deep dive into the evolution of SaaS for the long-term investor.

The current state of the software market is often described by self-proclaimed gurus as either an impending graveyard or a golden age of infinite productivity. It is truly impressive how many individuals possess a direct line to the divine, enabling them to predict with absolute certainty exactly which year Salesforce will become obsolete or precisely when a single AI agent will replace a thousand accountants. In the real world, where capital is actually at risk, such binary thinking is a luxury that investors cannot afford. The reality of the transition from traditional Software-as-a-Service (SaaS) to an agentic AI ecosystem is messy, non-linear, and governed by technical frictions that do not care about marketing slides.

Rationality suggests that since no one—including the CEOs of the largest tech companies—knows the final form of this disruption, the only sensible approach is to think in terms of probabilities and scenarios. To claim that AI is “just a bubble” is to ignore the large number of organisations already trying to use agents (and the reflexivity implications involved). Conversely, to claim that “software is dead” is to ignore the fundamental requirement for deterministic systems in any regulated business environment.

These notes provide a framework (just mine, you can think of your own) for understanding these shifting sands, utilizing the scenario models to map out the likely paths for the software sector through 2026 and beyond.

Welcome to Edelweiss Capital Research! If you are new here, join us to receive investment analyses, economic pills, and investing frameworks by subscribing below:

“The Parasol” – Francisco de Goya (1777)

---

🚀 New Project
For those who don’t know yet, let me remind you that together with Leandro @Invesquotes (Best Anchor Stock), we’ve launched a new podcast.

🎙️ New episodes every Tuesday at 18h in all platforms. For now, in Spanish only.
Subscribe/follow if you like it and share with anyone who might be interested.

I highly recommend our last two podcast episodes. We sat down with an InPractise analyst to dive deep into John Deere’s industrial moat—it’s a fascinating look at their tech evolution. We also explored the hard discounter model, specifically the disruptive new entrants seeing astonishing growth across emerging markets.

---

The Framework of Uncertainty: Scenarios and Structures

In a landscape defined by rapid technological shifts, the primary distinction that investors must make is between “Platform SaaS” (incumbents like ServiceNow or Oracle) and “Standalone/Non-integrated SaaS” (niche tools with limited data gravity).

I like the structure of Bain’s analysis here. It neatly evaluates these archetypes under different AI scenarios, and I’ve performed the public service of filtering out the fluff so we can get straight to the point.

The transition is moving the industry from a “Human + App” model to an “AI Agent + API” model. However, this transition is not happening at the same speed for every workflow. Some tasks are “perfect food for AI”—routine, rules-based, and digitally contained—while others are protected by a “Deterministic Bunker” of regulatory complexity and the need for human accountability.

To think about it, investors must abandon the hope for a single “correct” prediction and instead prepare for a range of outcomes, each assigned a probability based on technical feasibility and economic incentives, and your own thoughts.

Scenario 1. AI Has No Effect on SaaS

Narrative:

In this scenario, the initial wave of AI enthusiasm hits a wall of technical and regulatory reality. The narrative follows a “wait and see” approach where AI remains a helpful assistant but never becomes the primary driver of work. Platform SaaS companies like Microsoft or Salesforce continue to deliver value through traditional product roadmaps, and companies continue to pay for features and “seats” much as they have for the last decade. AI is viewed as an interesting research curiosity that provides marginal productivity gains but lacks the reliability for mission-critical business impact.

For non-integrated or niche SaaS tools, the world remains largely unchanged. Small productivity apps grow slowly, selling to individual users or small teams who encounter them through traditional marketing. The cloud/SaaS model, characterized by predictable subscriptions and recurring revenue, endures without the structural collapse many fear. This scenario assumes that “AI Hallucinations” and security risks prove too difficult for enterprises to manage at scale, leading to a “Flight to Determinism” where CIOs return to the safety of hard-coded software.

Implications (Pricing, TAM, Strategy, etc.)

(No need to develop)

Contra Argument

AI is already driving major changes. In reality, AI adoption is surging enterprise wide. Even Avenir notes “AI is a generational disruption” , so zero impact seems implausible. Historically, analogous scenarios (e.g. cloud vs. on prem) show incumbents eventually adopt new tech to stay competitive. Completely ignoring AI is unlikely unless the technology physically stalls or faces a global regulatory ban, both of which currently seem implausible given the competitive "arms race" between major tech powers.

Probability

~1-5%

Scenario 2. AI Enhances SaaS

Narrative:

In this scenario, AI acts as a massive tailwind for the existing software giants. AI features are seamlessly built into established platforms—think Salesforce Einstein, ServiceNow’s AI workflows, or Office 365 Copilot. Instead of replacing the software, AI acts as a “super-helper” that augments dashboards, automates mundane data entry, and provides predictive insights. The enterprise continues to use the core SaaS user interface and underlying processes, but the value of that software increases as it makes the human user twice as productive. These platforms deepen their moat by becoming a “System of Context,” embedding AI so deeply into the data flow that the platform becomes stickier than ever.

For niche SaaS, AI is a double-edged sword. On one hand, it’s a utility booster: a legal workflow app adds an AI assistant for document review, or a code-review tool allows a small team to manage a massive codebase. By amplifying individual output, AI turns ‘single-seat’ tools into ‘enterprise-grade’ engines.

However, there is a predator in the water. Historically, niche players survived because building specialized features was too expensive for the platform giants. If you built a clever integration inside the Salesforce ecosystem, you were relatively safe because Salesforce didn’t have the bandwidth to clone every ‘papercut’ solution.

Companies used to call these 'Best-of-Breed' solutions. Now, unless they have a proprietary data moat, we might as well call them 'Salesforce’s Unpaid R&D Department.' They spend the capital to prove the use case, and the platform uses AI to build the clone for pennies on the dollar.

Implications (Pricing, TAM, Strategy, etc.)

Pricing models in this scenario begin to evolve toward outcomes or usage. The traditional per-seat model faces pressure because if AI enables one person to do the work of two, the total number of seats a customer needs might decrease. To counter this “seat erosion,” SaaS firms must introduce new metrics, such as “tasks completed” or “results achieved,” to ensure that providing more value does not result in less revenue.

The TAM expands significantly as AI opens new budget lines; current projections suggest AI budgets will represent 8–12% of total IT spend by 2026. Furthermore, AI makes complex software accessible to smaller players; a $50,000 enterprise tool with a high onboarding cost can serve the SMB market via AI-driven self-service, implying a potential 10x TAM expansion for vertical players. Strategically, incumbents lean heavily into AI R&D and may “play dirty” by limiting API access to AI-native startups that try to disintermediate them.

Contra Argument

The risk in this scenario is over-optimism. AI projects frequently stumble on poor data quality and integration friction. Agents can make catastrophic errors if the underlying context is "messy". Enterprises may insist on rigid "guardrails" and human oversight for mission-critical roles, which could slow the pace of change significantly. Some CIOs recall the "chatbot revolution" of 2016 that failed to deliver, and critics point out that a well-designed clicking UI is often faster and more understandable than "chatting" with a probabilistic bot. If AI proves unreliable, SaaS growth will remain driven by existing features rather than new intelligence.

Probability

~30-40%

Scenario 3. AI Outshines SaaS

Narrative:

In this more radical vision, AI agents become the primary “users” of enterprise applications. A company’s internal AI system logs into Salesforce or ServiceNow to manage workflows autonomously, and the traditional user interface (UI) is sidelined. In this world, SaaS vendors are reduced to the “back-end engines”—the data stores and function libraries that agents rely on. Branding and feature differentiation through UX largely vanish because agents can hook into any data source via APIs.

Platform players attempt to protect their “System of Record” status, aiming to supply the specific data and semantic logic that agents need to function. However, even these giants may lose their front-end prominence. For non-integrated SaaS, the effect is even more stark. Small, single-purpose apps become interchangeable components. An AI might use the SaaS under the hood without the human user ever knowing it. Businesses may choose to pick only a few “AI orchestrator platforms” rather than subscribing to dozens of individual mini-SaaS tools.

Implications (Pricing, TAM, Strategy, etc.)

Business models shift decisively toward platform fees, integration fees, or usage-based API billing. If the human UI is no longer used, the “per-seat” license dies completely. Customers pay for data capacity or the number of transactions processed by the agents. Valuations for many firms likely compress, as features that become “AI accessible” tend to trend toward $0 pricing (commoditization).

While the TAM for software vendors might shrink as work is reallocated to the agent layer, the TAM for data platforms and AI orchestration services could rise. Strategy must pivot toward protecting “data lock-in” by offering agent-friendly APIs and rich metadata. Incumbents like Salesforce or ServiceNow may try to transform into the backbone of these agentic workflows, charging more for the “intelligence” and “integration” and less for the “seats”.

Contra Argument

The primary counter-argument is that “UI still matters.” For many tasks, clicking through a well-designed interface is more transparent and faster than “asking” an AI agent. Human-in-the-loop requirements and audit needs limit how far agents can go without constant oversight. Furthermore, the “ultimate problem of data quality” remains; AI agents struggle with fragmented or dirty data, often requiring they “call back” to human operators or traditional systems for verification. Reliable integration is incredibly difficult, and the complete dethroning of established SaaS platforms is not guaranteed in the near term.

Probability

~20-30%

Scenario 4. AI Cannibalizes SaaS

Narrative:

This is a more aggressive form of the “headless” scenario. Here, SaaS platforms are stripped down to their barest functions as infrastructure (storage, compute, and basic APIs). Even enterprise staples like CRMs and ERPs are relegated to being silent workhorses; their business logic is either opened up or re-implemented by AI. Essentially, each SaaS offering turns into “just a database or endpoint for agents”.

The market becomes layered: hyperscale AI providers (like OpenAI, Gemini, Gork or Anthropic) sit at the top, while commoditized cloud/SaaS sit at the bottom. Organizations pay primarily for compute and data, not for specific software features. For incumbents, the only remaining value is proprietary data or industry-specific compliance knowledge. Niche tools lose almost all customer interaction; any function they provide can be generated directly by a generic AI or by chaining together free APIs. For example, instead of subscribing to a chatbot service, a company uses a generic GPT agent that replicates that logic in-house.

Implications (Pricing, TAM, Strategy, etc.)

Pricing becomes almost entirely usage-based, with customers paying per API call or per gigabyte of data processed. Valuations will be based on the scale of data and the depth of integration, but profit margins will likely fall as software features are commoditized. The TAM for standalone SaaS collapses as many companies are either shut down or acquired at low multiples. Incumbents might attempt to survive by offering themselves as “private AI clouds”—selling a complete bundle of data, compute, and AI tailored to a specific enterprise. Strategic focus shifts toward “trust, privacy, and compliance” as the only remaining differentiated services.

Contra Argument

The total elimination of software layers is unlikely. Even if AI can generate code or perform tasks, businesses still need security, governance, and audit trails—roles that traditional software is specifically built to provide. Furthermore, the marketplace does not collapse instantly; customers may deeply mistrust “free” agents due to privacy or accuracy concerns. The “AI as plumbing” scenario also assumes that compute costs will remain cheap forever; if energy costs or regulations change, established software functions regain their value as efficient, specialized tools. Many businesses will cling to established workflows to avoid the risk of disruption.

Probability

~15-25%

Scenario 5. Spending Compresses (The Singularity)

Narrative:

This is the extreme “limit case.” In this vision, traditional applications disappear entirely. All workflows are handled by integrated AI networks. If a human wants to “do task X,” they tell a super-agent, which then invokes whatever code or data it needs internally. There are no longer distinct CRM or ERP tools, only integrated AI processes. Setting up emails, managing spreadsheets, or filing taxes—all are handled by a single, unified intelligence layer. “SaaS” as a market category vanishes, replaced by a few massive AI providers and the infrastructure companies that power them.

Implications (Pricing, TAM, Strategy, etc.)

(No need to develop)

Contra Argument

Complexity is the ultimate barrier. Even the most ardent AI advocates admit that neither SaaS nor AI can solve every problem. Highly specialized, safety-critical, or creative tasks will “always”? require human oversight or explicit, deterministic logic. Wholly abdicating mission-critical business decisions to a “black box” AI is far-fetched. History suggests that technology rarely completely replaces old tools overnight; even the move from on-premise to the cloud has taken decades and is still not 100% complete.

Probability

~1-5%

The Technical Reality: Probabilistic vs. Deterministic Intelligence

One of the most frequent errors in the “Software is Dead” narrative is a misunderstanding of how Large Language Models (LLMs) actually work. LLMs are, by their very nature, probabilistic and not deterministic. They are designed to be creative and to predict the next most likely word in a sequence; they are not designed to be compliant or to strictly follow a rigid set of rules.

In an enterprise environment, “creativity” is often a liability. You do not want your payroll software to be “creative” with tax withholdings. You do not want your security software to “hallucinate” a new firewall rule. Because LLMs are probabilistic, they can skip critical authentication steps or miss required compliance checks even with a detailed prompt. This is why the “Deterministic Bunker”—the hard-coded rules inside an ERP or CRM—remains the source of truth.

The most successful “agentic” systems being built today are not pure AI; they are hybrid systems that “wrap” non-deterministic models in deterministic infrastructure. If a task has known inputs and clear rules, a smart developer does not let the AI “decide” what to do; they use a hard-coded script. The AI is only used for the parts of the workflow that require judgment or natural language synthesis. This technical friction acts as a massive shield for incumbents who already own the “rules” and the “logic” of the industry.

The Battle for the “System of Context”

Avenir’s research suggests that the future of software is a choice between accepting maturity (financializing) or embracing AI to become a “System of Context”. Historically, SaaS providers were “Systems of Record”—they were the passive databases where companies stored their data. In the AI era, being a mere record is not enough. The high-value software of 2026 is a “System of Context”.

Avenir’s research

A System of Context does more than store data; it understands the “messy middle” of work. It ingests live conversations, emails, and unstructured documents to synthesize them into a “live pulse” of the business. While a System of Record tells you that a customer bought a product, a System of Context understands why they bought it by analyzing their sentiment during a sales call and comparing it to market trends.

The Valuation Mirage: Lower Prices vs. Cheap Stocks

It is tempting to look at a 40% or 50% drop in a software stock and conclude that it is “cheap.” However, a clinical assessment of the 2025-2026 data suggests caution. The median EV/Revenue multiple has compressed to 5.1x, which is a decade-low, but it is important to remember how “crazily expensive” these stocks were in 2021.

Avenir’s research

Furthermore, many SaaS firms still utilize Stock-Based Compensation (SBC) at levels that are, quite frankly, unacceptable to a rational owner.

Avenir’s research

Final thoughts

The debate isn’t about whether AI will ‘save’ or ‘kill’ SaaS—it’s about which companies possess the DNA to evolve. AI is either a tailwind or an adversary, depending entirely on strategy. We must watch for ‘AI-native’ platforms pushing into the system-of-record territory (as Avenir warns), while legacy SaaS players scramble to embed intelligence before their seats become redundant.

Expect chaos during this transition. We will see massive inefficiencies as companies either bankrupt themselves chasing AI gimmicks or starve their own innovation through caution. Our investment thesis leans into those playing offense—rewriting their pricing and re-engineering their products—while remaining wary of those clinging to the wreckage of a pre-AI world. It is an asymmetric battlefield: incumbents have the scale, but the agile have the speed.

As always, the future will be dictated by the best—the most innovative organizations that seize the right to create their own destiny. I’m looking for cultures built on relentless work and a capitalist spirit, led by entrepreneurial founders who go to work with a knife between their teeth.

What do you think? Now it is your turn! ;)

If you enjoyed this piece, please give it a like and share!

Thanks for reading Edelweiss Capital Research! Subscribe for free to receive new posts and support our work.

If you want to stay in touch with more frequent economic/investing-related content, give us a follow on Twitter @Edelweiss_Cap. We are happy to receive suggestions on how we can improve our work.