The start of 2026 has been particularly harsh for software sector stocks. The market, gripped by panic, has indiscriminately punished companies that were once untouchable. The IGV ETF, which groups major software companies, fell nearly 15% in January, one of the worst starts to a year on record. But this movement does not reflect a real crisis in the sector, but rather fear of a question that no one knows how to answer with certainty: what will agentic artificial intelligence do to the software business model?

The answer, however, is not as apocalyptic as some predict. Agentic AI will not kill SaaS, but rather separate the winners from the rest. To understand this, we must observe how the market is beginning to bifurcate software into two very different categories.

Two Types of Software, Two Different Futures

On one hand, we have interface-centric software, where the UI is the product and value depends on a human navigating screens and configuring rules. In a world where agents can execute tasks directly through APIs, this type of software risks losing much of its value. As Romain Boboe points out, there is an "interface tax": the more a product's value depends on manual interaction, the more friction it generates in an era that rewards autonomy.

On the other hand, there are cloud-native platforms, designed from the ground up to be operated by both humans and machines. They are API-first, data-centric, and built for orchestration and integration. In the age of agents, this type of software not only does not depreciate but multiplies its value. Sustainable advantage no longer comes from menus and buttons, but from data architecture, semantics, workflows, and connectivity.

Why AI Cannot (and Should Not) Replace a Well-Designed SaaS

There is a simplistic narrative gaining traction: "AI will generate the application" and enterprise SaaS will become optional. But the reality of critical systems is very different. Trying to replace a mature SaaS with agent-generated systems often increases cost and risk, and slows down the business. Let's look at three underlying reasons.

1. Higher Runtime Costs

AI reduces development costs but increases execution costs. Traditional software is cheap to operate because it is deterministic computation on CPUs. A user clicking "export report" is predictable and economical. In contrast, agentic AI consumes inference on GPUs: each interaction can be orders of magnitude more expensive than a traditional query. "AI-native" SaaS platforms thus become the optimization layer that keeps deterministic systems for what must be fast and cheap (transactions, permissions, storage) and introduces probabilistic AI only where it creates measurable value. In the agentic era, SaaS is not replaced by AI; it becomes the platform that controls the cost of intelligence.

2. More Risk and Higher Premium on Service

The "Service" in SaaS is often undervalued by those who think AI will replace everything. But for publicly traded companies, regulated sectors, and governments, service is essential. Organizations need commitments on availability, security, compliance, data governance, support, and accountability. If an agent generates an application for you, who do you call at 3 AM when it fails? Who answers to the regulator if data is leaked? When code becomes cheap, risk becomes expensive, and trust becomes more valuable. As we saw in the analysis of V-Valley, security and governance are prerequisites in any modern architecture.

3. Less Time to Market: Domain Depth Is Worth More Than Generated Code

The vision of "AI does everything" assumes that writing code is the bulk of a SaaS's value. In mature enterprise systems, that is rarely true. In a warehouse management system (WMS), perhaps 20% of the value is code; the other 80% is applied domain knowledge: edge cases that are only learned after seeing them in thousands of production environments. For example, how do you handle a partial receipt when the shipping notice says 1,000 units, but the physical count is 997 and 14 arrive damaged? Teams that build and operate enterprise SaaS have experienced these scenarios repeatedly. That knowledge is embedded in the product, which is why commercial platforms can work from day one of go-live. Custom development, even accelerated by AI, still needs years to discover what you didn't know you didn't know. As we pointed out in our article on dynamic workflows with Claude Code, agent autonomy does not replace domain knowledge.

Rethinking the Value of SaaS in the Age of Agents

When the dust settles, AI will not indiscriminately destroy software value. It will revalue software based on architectural realities, rewarding SaaS providers that deliver reliability, governance, accountability, and deployment speed, reinforced by the power of AI. AI will not be the end of SaaS, but the moment when the market separates the winners from the rest. And in that separation, domain depth, data architecture, and the ability to orchestrate inference costs will be key.

The author of this analysis is Eric Clark, President and CEO of Manhattan Associates, a position he assumed in February 2025. With a career that includes leadership roles at ServiceNow, Dell, Hewlett Packard Enterprise, and NTT Data, Clark brings a privileged perspective on how agentic AI is redefining enterprise software.

Original source: ComputerWorld. Analysis and adaptation by ForgeNEX.