“It is the most illogical thing in the world,” Nvidia CEO Jensen Huang said this week, bluntly dismissing the theory that AI will make traditional software obsolete. The remark landed in the middle of a brutal selloff across software stocks — a rout driven not by fundamentals alone, but by a suddenly viral idea: that agents and plug-ins can automate whole swaths of business software.
The market reaction has been dramatic. Investors hammered shares of incumbents from Salesforce to ServiceNow as new AI capabilities, particularly agentic tools and third‑party plug‑ins, promised to automate tasks once handled by specialized applications. Anthropic’s rollout of Claude Cowork plug‑ins — which stitch together workflows across legal, sales and analytics — catalyzed a fresh wave of selling, feeding a narrative that software vendors’ long-term value is at risk.
Two competing narratives
On one side sits fear: AI startups and model builders are moving fast to layer generalized agents on top of enterprise data, and some of those agents can perform tasks that used to require multiple software subscriptions. Traders see profit pools at risk and are repricing companies accordingly. A Deutsche Bank strategist captured the mood when he described a shift from “AI euphoria” to a period of differentiation — markets are trying to answer which companies will survive and which will be hollowed out.
On the other side is Huang’s counterargument: tools beget tools. He argues AI will rely on existing platforms to execute, not erase them. That’s not just spin from the maker of a chip that fuels many AI models — it’s a genuine strategic view. Many enterprise platforms already act as the plumbing for data, identity, security and integrations that agents need. Without that plumbing, agentic promises are fragile.
Why the truth is probably in the middle
Reality will likely look less like apocalypse and more like reconfiguration. Some categories — narrow workflow automation, template‑driven tasks, simple data aggregation — are most vulnerable. Others, where deep integrations, compliance, or industry‑specific workflows matter, will be harder and more valuable to displace.
Take Microsoft: it’s a cloud and software stack plus an AI partner to many enterprises, and it’s in the crosshairs of both narratives. The company is rapidly building in-house models and tools to host AI functionality, a reminder that big software players aren’t passive victims. Microsoft’s own moves on AI models and tooling underline how incumbents can evolve; see recent developments around its image model and tooling efforts in-house Microsoft Unveils MAI‑Image‑1.
Meanwhile, Google and others are adding agentic features to mainstream products, showing how big platforms will also compete to embed AI where users already live. Google’s experiments with agentic booking and task automation are a clear example of a different route to the same future Google’s AI Mode Adds Agentic Booking.
What investors and customers should watch
- Differentiation: Not all software products are equal. Platform companies that control data flows, identity, and compliance are harder to replace.
- Integration depth: Agents need hooks — APIs, connectors, admin controls — and that often means working with existing software rather than ripping it out.
- Monetization shifts: Expect pricing models to change. Some vendors may layer agentic features as premiums; others could see pressure on legacy subscription margins.
Some public companies cited by bullish analysts — Microsoft, Shopify, Figma — are pushing AI into their stacks and highlighting how incumbents can adapt. Shopify, for example, is baking generative features into merchant tools; Figma has been integrating AI into design workflows. And while those moves don’t immunize firms from competition, they do make a wholesale “AI replaces software” narrative less tidy.
Short‑term pain, long‑term competition
Markets are punishing uncertainty. When Anthropic’s plug‑ins hit the wires, investors sold first and asked questions later. But disruption and elimination are different. New AI agents will create winners — and losers — while also creating new opportunities for existing platforms to become the scaffolding that agents rely on.
If you’re watching the sector: expect more volatility as capabilities proliferate and as incumbents race to embed models, build integrations, and reprice their offerings. The debate Jensen Huang stoked is useful because it reframes the discussion: the question isn’t whether AI will change software — it’s which companies will turn change into advantage.