Is It the End of Software?

Executive Summary

Recent market moves have forced a sharper question onto the software sector. Investors are no longer asking only whether AI will help software companies move faster. They are asking whether advances in AI change the basis on which software earns durable value in the first place. Reuters reported that the S&P 500 Software & Services Index had already fallen 25.5% in 2026 by early April, and that Anthropic's latest model release reignited disruption fears across U.S. software stocks.

That shift does not mean software is disappearing. It means software is being repriced. AI is changing both how software is built and how work is executed through software. If building software becomes easier, and if more work can be carried out through natural-language instruction and agentic systems, then some older forms of software defensibility become less sufficient on their own.

Not all software is equally exposed. Analysis summarized by Business Insider from AlixPartners suggests that categories with replaceable features and thinner moats look more vulnerable, while software tied to proprietary data, vertical depth, and regulated operating environments may prove more resilient.

Payroll becomes important in that context not because it is immune to AI, but because it is a revealing test case. Payroll is clearly exposed to AI in the sense that more of its surrounding knowledge work can now be systemized. But payroll also depends on permission control, sensitive data handling, deterministic execution, auditability, and trust. It sits directly on the boundary between software that may weaken in the AI era and software that may become more essential because of it.

Software is not disappearing. It is being repriced.
AI is changing both how software is built and how work is executed through software.

Why is SaaS suddenly being repriced?

For most of the last decade, enterprise software enjoyed a privileged position in both capital markets and operating budgets. Recurring revenue, embedded workflows, switching costs, and durable retention made SaaS look like one of the most defensible business models in technology.

That consensus is now under pressure. Reuters reported on April 9 that U.S. software stocks sold off sharply after Anthropic's latest model release reignited fears that AI could disrupt large parts of the traditional software industry. By that point, the S&P 500 Software & Services Index was already down 25.5% in 2026. In late February, Reuters also reported that Workday had fallen to a more than five-year low after issuing a weak sales forecast, with AI competitiveness concerns explicitly part of the investor discussion.

The important point is not that the market has decided software is dead. It is that the market has stopped assuming software is automatically safe. AI is no longer being discussed only as a productivity tailwind for software vendors. It is now part of the question of whether some software categories still have durable protection at all.

What is AI really changing about software?

The most important shift is deeper than product features. Software was never fundamentally about code, screens, or subscription packaging. Those were delivery forms. At its core, software has always been a method for automating non-physical work: structuring inputs, applying rules, coordinating actions, preserving records, and making complicated operations repeatable at scale.

AI now affects that system from both directions. First, it changes how software is built. Many software products are becoming easier, faster, and cheaper to develop, which lowers the scarcity of software creation itself. Second, it changes how work is executed through software. Tasks that once required dedicated workflows, fixed application logic, or specialized application structures can increasingly be initiated and carried out through natural-language instruction and agentic systems.

That does not make software irrelevant. But it does make older forms of defensibility less sufficient on their own. If software becomes easier to build, and if parts of work can be executed through agents rather than only through conventional applications, then durable value shifts away from form and toward what remains difficult to replace: governed access to important data, embedded domain logic, controlled execution, reproducibility, accountability, and trust.

What does this mean for Salesforce, Workday, and ServiceNow?

These companies should not be read as collapse stories. They are better understood as transition stories. The relevant question is how effectively large, well-resourced software vendors can adapt to an environment in which AI changes both the economics of software creation and the mechanics of software-mediated work.

Their scale, installed base, data assets, and product breadth give them real capacity to evolve. At the same time, legacy architectures, broad product surfaces, pricing structures, and organizational complexity may make that evolution harder than it is for more focused next-generation entrants. The market reaction is therefore best read not as proof of decline, but as pressure to explain a credible AI-era transition path.

So, what kinds of software may prove more durable?

Not all software is equally exposed. Recent AlixPartners analysis, highlighted by Business Insider, suggests a differentiated split inside enterprise software. Marketing automation, productivity tools, CRM add-ons, and analytics products appear especially vulnerable because their features are relatively replaceable and easier for AI-native competitors to absorb. By contrast, software used in regulated domains and software anchored in proprietary data or vertical specialization appears more resilient.

This distinction makes intuitive sense. Software becomes more durable when it does more than provide efficiency around routine work. It becomes more durable when it governs consequential work. That usually means some combination of the following: access to sensitive or hard-to-replace data; embedded domain-specific logic; controlled execution under real-world rules; legal, financial, or regulatory consequences tied to output; and the need for records that can later be audited, reproduced, and defended.

In such environments, AI does not automatically eliminate the need for software. In many cases, it raises the value of software that can combine intelligence with control.

Why do stronger models make control more important?

This is the missing bridge in much of the SaaS debate. The common assumption is that stronger AI means software matters less because more tasks can be handled directly by models or agents. But in real operating environments, stronger AI can make control more important, not less.

The more capable AI becomes, the more important it becomes to know what data it is allowed to access, under what permissions it is operating, what rules it is applying, what actions are deterministic and final, what records are preserved for later review, and where accountability still sits. Automation capacity is not the same thing as operational legitimacy.

This is especially true where outputs carry legal, financial, employment, or compliance consequences. In those settings, it is not enough for AI to produce something plausible. The output must be reproducible where necessary, reviewable by humans, explainable in context, and tied to a system that preserves trust.

Why does payroll deserve separate attention?

This note did not begin as a payroll note. It began as a question about software: if AI is forcing a repricing of SaaS, then what kinds of software may actually emerge stronger?

Payroll is one of the clearest examples because it is exposed in both directions at once. On one hand, payroll is clearly exposed to AI. Large parts of the surrounding work in payroll have historically remained outside the system: employee questions, interpretation of edge cases, exception handling, reconciliation support, audit explanation, and operational support across jurisdictions and stakeholders. AI is the first technology wave that can realistically help bring more of that knowledge work into software.

On the other hand, payroll is not just another automation category. Payroll governs highly sensitive employee and financial data. It operates under permission boundaries. It sits inside statutory obligations and deadline-bound execution. Its outputs must be correct, reproducible, explainable, and auditable. That makes payroll less like a generic software workflow and more like governed operational infrastructure.

ADP's own public messaging points in this direction. In February, ADP said APAC organisations are increasingly using AI to run payroll with leaner teams, improve accuracy, and manage growing operational complexity. That is not evidence that payroll is immune to AI. It is evidence that payroll is likely to evolve with AI, and that the real question is what kind of payroll software that evolution will favor.

What would AI-era payroll software actually require?

The phrase AI-native payroll can be overused, but the underlying requirement is real. In payroll, the meaningful distinction is not cosmetic. It is architectural. The next generation of payroll software will likely need to do two things at once: absorb more of payroll's surrounding knowledge work, and preserve the controls that payroll cannot do without.

That means, at minimum, permission-aware access to sensitive employee and payroll data; embedded statutory and operational logic; clear separation between reasoning support and final execution; reproducible calculation and filing outcomes; preserved auditability and system-of-record integrity; and explicit accountability around what AI can suggest, what it can prepare, and what the system can actually execute.

This is why the principle often described as probabilistic where useful, deterministic where it matters is so relevant here. Traditional payroll software often offered deterministic execution without adequately absorbing the surrounding knowledge layer. Generic AI tools offer flexible reasoning without the governance layer payroll requires. The more durable payroll systems may be those that combine both.

The real conclusion

The “end of SaaS” thesis is too crude. Software is not disappearing. It is being repriced.

Software was never merely a codebase or an interface format. It was always a method for automating non-physical work. AI is now changing both how that software is built and how that work is executed. That creates pressure on some software categories, but it also clarifies what kinds of software remain essential.

The categories most likely to matter are not simply the ones with the most visible AI features. They are the ones that combine intelligence with what remains hard to replace: governed data access, embedded domain logic, controlled execution, reproducibility, accountability, and trust.

Payroll deserves attention because it sits directly on that boundary. It is clearly exposed to AI in the sense that more of its surrounding work can now be systemized. But it is also one of the clearest examples of software whose deeper value lies in governance, permissions, compliance, and trust. That is why payroll is not just a side topic in the SaaS debate. It is one of the most revealing categories through which to understand it.

References and source notes

1. [1] Reuters. “US software stocks slump on renewed AI disruption jitters.” 9 Apr 2026.
2. [2] Reuters. “Workday hits over five-year low as sluggish sales forecast sparks AI disruption fears.” 25 Feb 2026.
3. [3] Business Insider. “A new scorecard shows which software companies will win or lose in AI.” Apr 2026 (summarising AlixPartners analysis).
4. [4] ADP Singapore. “AI-Enabled Efficiency to Shape Payroll Transformation Across APAC in 2026.” 25 Feb 2026.
5. Source note. The market observations in this note are grounded in the sources above. The analytical conclusions, framing, and category interpretations are original synthesis.