SaaS Faces a Sorting, Not an Apocalypse, From AI Agents

Todd McKinnon described the current shift to AI agents as larger than any technology transition since the internet, “possibly bigger.” For Okta, his claim is not simply that AI will create another software cycle. It is that agents increase the importance of identity because they add automated actors that need governed access across companies’ existing software environments.

Okta, McKinnon said, was founded during the move to cloud computing. The AI shift now creates a comparable opening, but with a different center of gravity: agents need an identity and connectivity layer if they are going to operate across enterprise systems. That is the premise behind “Okta for AI Agents,” which he described as a product for managing that layer inside companies.

The differentiation, in McKinnon’s account, is focus. Many vendors want to be the governance or protection layer for AI. Okta is emphasizing the connections themselves: the links that let agents reach the information and tools required to be useful. Those links are also where risk concentrates. The more an agent can reach, he said, the more potential there is for information leakage, an agent going rogue, or a need to shut it down quickly.

“Connectivity” is therefore not neutral plumbing in McKinnon’s framing. It is where usefulness and security collide. McKinnon said Okta has a blueprint for connecting agents securely, and that this is resonating with customers.

Ed Ludlow pressed McKinnon on the market reaction to Jensen Huang’s argument that the negative narrative around software is “complete nonsense” and that agents will become end users of legacy software. Bloomberg’s five-day chart shown during the segment put Okta shares at 123.82, up 29.10, or 30.72%, over the displayed period from May 29 to June 4.

McKinnon did not treat Huang’s comments as simply good or bad for software stocks. He placed them inside a broader technology-stack transition. In his view, AI is forcing every layer to be reimagined: hardware, infrastructure, platforms, development tools, and applications. Every vendor, he said, has to reinvent itself around agentic capabilities and automation.

His added point was that this cycle is not limited to making existing software products more automated. McKinnon argued that AI creates “a new layer” above the stack: digital work. That layer is incremental to the reinvention of current software categories. Vendors still have to modernize their own products, but there is also a separate market forming around automated work performed by agents.

That distinction underpins his view of Okta’s opportunity. He described two drivers for the business: the renewed importance of identity as companies rebuild for AI, and the emergence of automated digital work, where Okta for AI Agents is meant to provide identity and secure connectivity.

McKinnon accepts that vendors must change. He rejects the idea that AI agents make software companies broadly obsolete.

Caroline Hyde asked what the AI shift means for Okta’s own workforce and whether the company will need as much human capital. McKinnon said employees will have to become more effective and more productive, using tools such as Claude Code and newer marketing and support tools. The standard, as he put it, is that people “have to deliver value.”

But when Hyde sharpened the question into whether Okta would keep the same number of people, McKinnon disagreed with what he said many people expect. In five years, he said, Okta will have “far more software engineers” than it has now.

His explanation was not that engineers will avoid productivity gains. It was that productivity gains will expand what the company can and should build. There is, in his account, more software demand than current organizations can satisfy, and automation increases the amount of capability worth building.

Ludlow pushed the agent discussion away from abstraction. His example was a canceled flight: today he may still need to call a human to resolve it, even inside a sophisticated organization. The agentic version would have access to the credit card, flight information, and relevant systems, and would simply act.

McKinnon said that example is exactly what he means by connections. The hard part is not only granting access, but securing it.

The barrier, in his view, is not primarily that models lack raw capability. He argued that model capability now “far surpasses” companies’ ability to connect systems and build practical agentic applications. The work ahead is making those connections real, usable, secure, and manageable.

Ludlow asked whether companies should be worried about how secure such transactions would be. McKinnon said they should be. Connecting sensitive data and automating action around it creates new risk. Okta’s claim is that enterprises need a pragmatic way to give agents more data and capability without losing control of how access is managed.

For Okta, the connection layer is where identity, access, security, and agent usefulness meet.

When Hyde asked directly about talk of a “SaaSpocalypse,” McKinnon did not hedge.

His objection was not that every SaaS company is safe. It was that the phrase misses the pattern he expects from a platform transition.

Some SaaS companies, he said, will fail to embed AI and agentic capabilities into their offerings. He compared that to the cloud transition, when some companies were left behind, became legacy vendors, and were “bought out by Oracle.”

But he also said many vendors will make the transition and thrive. In McKinnon’s view, the shift is not an apocalypse for SaaS as a model, but a sorting process. Companies that adapt their products for AI agents and automated work can benefit from the shift. Companies that do not may become legacy software.