Todd McKinnon has spent the last 15 years building Okta into the dominant force in digital identity management. Now he’s making the biggest strategic bet of his tenure — and it hinges on a simple but provocative idea: AI agents need identities just like people do.
In a recent conversation on The Verge’s Decoder podcast, McKinnon laid out his vision for a world where billions of autonomous software agents interact with enterprise systems, make decisions, and execute tasks — all without a human clicking a button. The question he’s asking isn’t whether that world is coming. It’s who will manage the trust layer underneath it.
His answer, unsurprisingly, is Okta.
But the argument is more nuanced than a CEO talking his own book. McKinnon is pointing at a structural gap in enterprise infrastructure that almost no one has filled. When a human logs into Salesforce, there’s a well-understood chain: username, password, multi-factor authentication, permissions. When an AI agent tries to do the same thing on behalf of a human — or on behalf of another agent — the existing frameworks break down. There’s no standard for how an agent proves who it is, what it’s allowed to do, or who’s accountable when it does something wrong.
That’s the opening McKinnon is driving Okta toward.
The Agent Identity Problem No One Solved Yet
The rise of agentic AI — autonomous software that can reason, plan, and take action across multiple systems — has moved from research curiosity to boardroom priority in roughly 18 months. OpenAI, Anthropic, Google, Microsoft, and Salesforce have all announced or shipped agent frameworks. McKinsey estimates that agentic AI could automate 60-70% of workers’ current tasks. Enterprises aren’t asking if they’ll deploy agents. They’re asking how fast.
But speed has exposed a fundamental problem. As McKinnon explained on Decoder, the identity layer for these agents barely exists. “The identity problem for agents is actually harder than it is for humans,” he said, noting that agents can spawn other agents, operate across organizational boundaries, and act with a degree of autonomy that doesn’t map neatly onto traditional role-based access controls.
Think about it this way. A sales rep uses an AI agent to pull customer data from a CRM, cross-reference it with financial records, draft a proposal, and send it for approval. That agent needs access to at least three different enterprise systems. Who granted that access? Does the agent inherit the sales rep’s permissions wholesale, or does it get a narrower set? What happens if the agent is compromised? What if it hallucinates a financial figure and sends the proposal before anyone reviews it?
These aren’t theoretical questions. They’re active headaches for CISOs at every Fortune 500 company deploying agent technology right now.
Okta’s response is a product called Auth for GenAI, which McKinnon described as purpose-built infrastructure for authenticating and authorizing AI agents. The product handles what Okta calls the “identity lifecycle” for agents — creation, authentication, scoping of permissions, and eventual deactivation. It also addresses a thorny sub-problem: agents acting across different organizations, where trust boundaries get complicated fast.
The company has also been working on what it calls the “Agent Service Account,” a new identity primitive designed specifically for non-human actors. Unlike traditional service accounts — which have been a security nightmare for decades, often holding static credentials with overly broad permissions — Okta’s version is designed to be dynamic, time-limited, and auditable.
McKinnon framed this as the natural evolution of Okta’s core business. The company started by managing human identities in the cloud. Then it expanded to workforce identity and customer identity. Now it’s adding machine and agent identity as a third pillar. “This is not a pivot,” he told The Verge. “This is the next chapter of the same story.”
Wall Street seems cautiously optimistic. Okta’s stock has recovered significantly from its 2022-2023 lows, though it still trades well below its pandemic-era highs. The company posted strong fiscal Q1 2025 results earlier this year, with revenue growth reaccelerating and margins improving. Analysts at firms including Morgan Stanley and Bernstein have flagged the agent identity opportunity as a potential growth catalyst, though most models don’t yet bake in significant revenue from it.
A Crowded Field, But an Empty Center
McKinnon isn’t the only executive who’s noticed the gap. Microsoft has been building agent identity features into its Entra ID platform. CrowdStrike has talked about non-human identity as a security vector. Startups like Astrix Security and Oasis Security have raised significant venture capital to tackle machine-to-machine identity specifically. And the big cloud providers — AWS, Google Cloud, Azure — all have their own IAM (identity and access management) stacks that are being extended for AI workloads.
So why does McKinnon think Okta has an edge?
Two reasons, primarily. First, Okta is identity-neutral in a way that the hyperscalers aren’t. An enterprise running workloads across AWS, Azure, and Google Cloud doesn’t want its identity layer locked into any one of them. Okta sits on top of all three, which gives it a natural position as the identity broker in multi-cloud environments — exactly the kind of environments where agents will operate.
Second, Okta already manages over 18,000 enterprise customers and processes billions of authentications per month. That installed base matters enormously. Enterprises don’t rip out their identity provider lightly. If Okta can extend its existing contracts to cover agent identity, the incremental revenue comes with low customer acquisition cost and high switching costs. It’s the classic platform expansion play.
But there are real risks. Okta’s 2022 security breach — in which attackers compromised an Okta support engineer’s credentials and accessed customer data — still lingers in the memory of security professionals. For a company whose entire value proposition is trust, that incident was deeply damaging. McKinnon has acknowledged it publicly and repeatedly, pointing to the security investments Okta has made since. But when you’re asking enterprises to trust you with the identity layer for autonomous AI agents, the bar for credibility is extraordinarily high.
There’s also the question of standards. The agent identity space is pre-standardization. There’s no equivalent of SAML or OAuth specifically designed for AI agents, though working groups at organizations like the OpenID Foundation are starting to discuss it. McKinnon told Decoder that Okta is participating in these efforts, but he also seemed comfortable moving ahead of formal standards — a strategy that could pay off if Okta’s approach becomes the de facto standard, or backfire if the industry coalesces around something different.
And then there’s the competitive timing issue. Microsoft’s Entra ID has massive distribution through its enterprise agreements. If Microsoft decides to bundle agent identity management into its existing security stack at no additional cost — a move it has made repeatedly in adjacent categories — Okta could find itself fighting the same bundling headwinds that have pressured other best-of-breed security vendors.
McKinnon seems aware of this. He argued on the podcast that identity is too critical and too complex to be an afterthought bundled into a larger platform. “You don’t want your identity managed by a company that also wants to sell you compute and storage and databases,” he said. It’s a compelling argument philosophically. Whether it holds commercially against Microsoft’s distribution machine is another matter entirely.
The broader context here is the rapid expansion of what security researchers call the “non-human identity” attack surface. A report from CSO Online noted that non-human identities — service accounts, API keys, tokens, bots — already outnumber human identities in most enterprises by a ratio of 45 to 1. AI agents will dramatically expand that ratio. Every agent, every sub-agent, every tool-use session represents a new identity that needs to be managed, monitored, and secured.
This is a massive market expansion opportunity. Gartner has estimated that by 2028, a third of enterprise software interactions will be handled by autonomous agents. If that forecast is even directionally correct, the identity management market — currently valued at roughly $20 billion — could double or triple in size as agent workloads scale.
McKinnon is betting that Okta can capture a disproportionate share of that growth. The company’s recent product announcements suggest it’s investing heavily: new APIs for agent authentication, partnerships with major AI model providers, and an expanded developer platform designed to make it easy to integrate Okta identity into agent frameworks built on LangChain, CrewAI, and similar toolkits.
The Trust Question That Underpins Everything
Strip away the product announcements and market sizing, and McKinnon’s thesis comes down to a single word: trust. In a world where AI agents are making consequential decisions — approving expenses, modifying customer records, executing trades, writing code — someone has to answer the question: “Who authorized this, and can we verify it?”
That question is fundamentally an identity question. And it’s one that gets harder, not easier, as agents become more capable.
Consider the emerging pattern of multi-agent systems, where one agent delegates tasks to other agents, which may in turn invoke tools or APIs from third parties. The chain of trust in such a system can be several layers deep. If something goes wrong — a data leak, an unauthorized transaction, a compliance violation — enterprises need to trace the entire chain back to its origin. That requires not just authentication at each step, but a persistent, auditable identity for every actor in the chain.
No one has built this at scale yet. Not Okta, not Microsoft, not anyone.
McKinnon’s argument is that Okta is better positioned than anyone to build it first. Maybe. The company has the domain expertise, the customer relationships, and the technical foundation. It also has the scars from its own security failures, which — if McKinnon is right that they’ve made the company more rigorous — could paradoxically be an advantage.
But the race is early. The standards are unwritten. The competitive dynamics are fluid. And the technology itself — agentic AI — is evolving so rapidly that today’s architectural assumptions could be obsolete in 18 months.
What’s clear is that the identity layer for AI agents isn’t optional. It’s the load-bearing wall of the entire agentic enterprise. Someone is going to own it. McKinnon has decided that someone will be Okta — and he’s reorganizing the company around that conviction.
Whether that conviction is visionary or premature will be one of the defining questions in enterprise technology over the next three years. For now, McKinnon is all in. And the stakes, for Okta and for the industry, couldn’t be higher.
WebProNews is an iEntry Publication