What does agentic AI mean for the future of software?

January was a brutal month for software stocks. Sentiment swung hard, and in many cases, it had less to do with fundamentals and more to do with a single question hanging over the sector, “What does agentic AI mean for the future of software?”

A recent William Blair equity research note* illustrates the mood well: the IGV software ETF was down roughly 15% in January, making it one of the toughest Januarys in the ETF’s history. The bigger point isn’t the exact number. It's what it represents: fear is compressing nuance, and markets are treating “software” as one bucket. But reality is splitting into two very different categories.

The market is penalizing “software” broadly, but software is bifurcating.

On one side: interface-centric, configuration-heavy software, where the UI is the product. In a world of autonomous workflows, that value is likely to erode. Agents will increasingly bypass screens, perform tasks via APIs, and collapse some categories where “clicking around” was the primary mode of work.

On the other: API-first, data-centric, cloud-native platforms. This class of software was built to be operated by machines as much as by humans, engineered for orchestration, integration, and scale. In the agentic era, that’s where value will concentrate and compound.

In other words, in a world of agents, durable advantage doesn’t come from menus and buttons.

It comes from data architecture, semantics, workflow, and connectivity.

Romain Boboe’s idea of an “interface tax” captures this well: the more your product’s value depends on a human navigating screens and configuring endless rules, the more friction you’re carrying into an era that rewards autonomy.

The missing point: AI can’t (and shouldn’t) replace well-engineered SaaS.

There’s a popular narrative that “AI will generate the app” and enterprise SaaS becomes optional.

That’s not how mission-critical systems work in the real world.

AI is powerful, and it will absolutely change how software is built and used. But in many enterprise domains, trying to replace well-engineered SaaS with agent-generated or agent-run “DIY systems” often increases cost and risk, and it slows the business down.

Three reasons why AI won't be the end of SaaS

Here are three reasons why AI won’t be the end of SaaS, and why the strongest SaaS organizations are uniquely positioned to make AI more valuable.

Higher run-time costs

AI will reduce development costs, even meaningfully. But it increases runtime costs, often significantly.

Traditional enterprise software is comparatively cheap to run because it’s largely deterministic compute on CPUs. A user clicking “export report” or running a query is computationally predictable and relatively inexpensive.

Agentic AI changes the economics.

Natural language Q&A, summarization, extraction, reasoning workflows, and multi-step agents burn inference compute, often on GPUs or other accelerators. Each interaction can be orders of magnitude more expensive than a traditional database query.

So, while “making software” may get cheaper, running intelligent software will be more costly and more complex.

This is exactly where AI-native SaaS platforms shine: they become the optimization layer between customers and chaos.

• Keep deterministic systems for what must be fast, cheap, and correct (transactions, permissions, storage, ledgers).
• Inject probabilistic AI only where it creates measurable value (assistance, discovery, synthesis, automation).
• Orchestrate the hybrid infrastructure, so customers don't have to become experts in GPU scheduling, prompt caching, retrieval pipelines, model routing, safety layers, and cost governance.

In other words, in the agentic era, SaaS isn’t replaced by AI; it becomes the platform that controls the cost of intelligence. 

Increased risk, and the premium value of “service”

The “Service” in SaaS is often overlooked by people who expect AI to take over enterprise solutions. But for public companies, regulated industries, government entities, and anyone concerned about operational risk, the “Service” is essential.

Modern enterprises need mission-critical commitments they can stand behind:

• Availability (and credible uptime guarantees)
• Security and compliance (SOC 2, HIPAA, ISO, auditability)
• Data governance (access controls, retention, lineage, residency)
• Support and operations (incident response, SLAs, training)
• Consistency over time (“it works tomorrow exactly as it worked today”)
• Accountability and trust (who owns failure when it matters?)

If an AI agent generates your application, who gets called at 3 a.m. when it fails? Who answers regulators if sensitive data leaks? Who can prove exactly why a system made a decision?

Operational accountability, governance, and reliability remain premium value, and they don’t go away in an agentic era. If anything, they become more important.

In other words, as code gets cheaper and more autonomous, risk becomes more expensive.

Reduced time to market: domain depth beats generated code

The “AI-does-everything” view often assumes code creation is the majority of the value in a SaaS solution. In mature, mission-critical enterprise systems, that’s rarely true.

In a system like a warehouse management platform, maybe 20% of the value is the code itself. The other 80% is applied domain knowledge - the countless edge cases and operational realities you only learn after you’ve seen them across thousands of live environments. For example:

• How do you handle a partial receipt when the ship notice says 1,000 units, but the physical count is 997, and 14 are damaged?
• What’s the downstream impact on allocation, replenishment triggers, and inventory valuation?
• How do cycle count discrepancies trigger different workflows depending on SKU velocity class, value thresholds, or time since last full inventory?

Teams that build and run enterprise SaaS have lived these scenarios repeatedly. That knowledge is embedded into the product, and that’s why commercial platforms can work on day one of go-live.

A custom build, even one accelerated by AI, still takes years to discover what you didn’t know you didn’t know.

In other words, infused AI makes modern SaaS faster and more efficient. It doesn’t make it optional.

Revisiting the value of SaaS

After the dust settles, AI isn’t going to kill software value indiscriminately.

It’s going to reprice software based on architectural realities, rewarding the SaaS providers who deliver the traditional value of enterprise software (reliability, governance, accountability, time-to-value) infused with the promise and power of AI.

AI won't be the end of SaaS. It will be the moment the market separates the winners from the rest.

*Arjun Bhatia et al., “Fear, Not Fundamentals: the Sentiment Problem and Why Software Is So Challenging Right Now,” William Blair Equity Research, Technology, Media, and Communications, February 2, 2026.