In early February, Wall Street erased $2 trillion in market value from the software sector, as markets repriced the “unbundling” of software in the AI era. Traders even created a name for it: the SaaSpocalypse. I felt a weird sense of déjà vu. Through my work at CareYaya in AI and healthcare innovation, we’ve been unbundling a different kind of legacy middleman for years: the traditional senior care agency. What I’ve learned is that when you remove the layers in a business that exist mainly to manage friction, the “moat” turns out to be switching costs and coordination costs, and both can collapse faster than anyone expects.

Now, I’m seeing the strangest shift in software more broadly. It’s not that AI can write code (although it can do that quite well!) It’s that AI is starting to buy code. Let me explain.

Other WRAL Top Stories

For the past two decades, the center of gravity in tech has been the human user: the admin logging into the dashboard, the manager approving a workflow, the employee clicking through tabs. We built entire empires around that click path.

But now, as autonomous AI agents get embedded into everyday work, the “user” increasingly looks like a piece of software with a budget, a set of permissions, and zero patience for your onboarding flow.

>> Tom Snyder: What the market knows about the future of software

Economist Ronald Coase gave us a clean lens for what is happening. Firms, he argued, exist because markets have friction. Finding a specialist, negotiating, enforcing, coordinating, verifying, all of that has a cost. When those transaction costs are high, it’s cheaper to hire employees and build inside the firm. When those transaction costs fall, the boundary of the firm moves. That 1937 insight is suddenly feeling like a live product roadmap.

AI agents are the transaction cost wrecking ball.

In the last year, we’ve watched major platforms explicitly reorganize around autonomous agents that act across systems, not just inside one app. Gartner is blunt about where this goes: task specific agents get embedded into a large share of enterprise apps fast. If the agent can discover a service, compare options, and call an API in seconds, then a big chunk of “selling software” becomes irrelevant.

This is why the old SaaS playbook suddenly feels shakier. The historic moat for many of these companies was switching costs. Data trapped in a proprietary model, workflows entangled with a vendor’s UI, and institutional habit. That moat is getting lower, because AI agents are becoming translators. They can map one schema to another, migrate objects, rewrite automations, and clean up the messy middle layer humans used to dread.

You can already see the downstream effects in how executives talk about headcount. Klarna’s CEO has said the company expects its workforce to shrink further by 2030 after major reductions, explicitly tying the trajectory to AI driven efficiency. Love it or hate it, leaders are betting that more workflows become machine operated, and fewer roles exist purely to “move information between systems."

Now bring this home to North Carolina, because our backyard may be the clearest place to see what an agentic economy actually means.

Our population is aging fast. By 2030, older adults are projected to outnumber children in North Carolina for the first time. Healthcare, notoriously, is also where systems of record have been the most fortified. But policy is pushing the opposite direction. The federal push against information blocking, and toward access and exchange of electronic health information, is a preview of what “data portability” looks like in a regulated market. Pair that with AI agents, and the old idea that a vendor “owns” a workflow rapidly starts to break down.

I think this is the opportunity for builders in the Triangle and beyond. The winners will not be the teams who rebuild yesterday’s dashboards with AI sprinkles, but those who make services legible to machines.

A practical checklist looks like this:

  • Make capabilities machine readable, not the way marketing copy used to be.
    • Price and permissions need to be computable at runtime.
      • Onboarding must be automatable, not a sales-led scavenger hunt.
        • Interoperability becomes your distribution channel.

          Standards are catching up to this agentic reality too. Anthropic’s Model Context Protocol push, and the move toward registries and shared tooling, is a signal that the ecosystem wants common rails for “agents talking to tools”.

          I also think there’s a moral question buried under all of this. If the buyer becomes an agent optimizing for speed and cost, we risk building a world that is brutally efficient and strangely indifferent. In elder care, in healthcare claims, in patient navigation, the edge cases are the human beings. The widow who is confused by a medical bill; the son who is trying to keep his job while coordinating dementia care for his father.

          So yes, software is changing shape. But the point through tech innovation should not be to make humans obsolete. The point is to make coordination cheaper, so that products and services can be more abundant and accessible for all.

          In an agentic world, I think the companies that win will be the ones that can be trusted by machines and still feel humane to the people living with the consequences.

          Neal K. Shah is the CEO of CareYaya Health Technologies. He is an NIH-funded Principal Investigator on the YayaGuide AI for Caregiver Training project that he started at Johns Hopkins, and the Co-Principal Investigator on the University of Pennsylvania-funded Counterforce Health project on artificial intelligence for health insurance denials. Neal also serves on North Carolina's Steering Committee on Aging.