Okta CEO Todd McKinnon on AI agents, identity security, and why enterprise software categories are up for grabs
Feb 12, 2026 with Todd McKinnon
Key Points
- Okta CEO Todd McKinnon says enterprises are deploying AI agents into production without proper access controls, creating immediate security exposure as agents connect to multiple data sources with excessive permissions.
- McKinnon predicts AI will expand total software creation so dramatically that both custom agentic systems and enterprise packaged software spending increase simultaneously over the next five years.
- Identity security, traditionally a modest cybersecurity category, could become the largest within a decade as AI disruption makes all enterprise software leadership positions vulnerable to category redefinition.
Summary
Todd McKinnon, co-founder and CEO of Okta, says enterprise boards are pushing AI agents into production faster than security teams can secure them. CEOs read about agents on social media, commit to the technology, and demand demos go live without proper access controls. The real risk is basic. Agents connect to multiple data sources with excessive permissions, and a few prompt injections expose customer data across the entire company. McKinnon points to live cases where agents hooked to multiple systems reveal customer information from different accounts through simple queries. This is fundamental access control breaking under agent proliferation, not an exotic edge case. Okta's role is to add visibility and control early, before companies face public breaches.
McKinnon makes a contrarian claim about the broader software market. In five years there will be more software engineers than today, not fewer, despite AI automation. The twist is that there will be far more software overall. Developers will be more productive and also more numerous. Companies will keep building in-house, particularly cross-silo orchestration that traditional vendors cannot deliver because their platforms are siloed by design. HR vendors cannot think end-to-end. Salesforce and SAP cannot either. Custom agentic systems will fill that gap. Enterprise packaged software spending will still rise despite the custom wave. Open-source equivalents of major platforms have existed for decades without displacing vendors. The scale of new software creation will fund both categories.
McKinnon argues that AI is disrupting the settled leadership of enterprise software categories. Five or ten years ago AWS seemed locked in as the infrastructure leader. Google, Oracle, and others are now plausible contenders. Identity, traditionally modest relative to firewalls, SOCs, and endpoint security, could become the largest cybersecurity category within a decade. ServiceNow and others see this shift and are acquiring accordingly. Every major category is vulnerable to disruption through redefinition itself, not just through startups. The most exciting new software will be agentic systems that cross silos, something humans excelled at but packaged vendors structurally cannot because they operate within their own data models and business rules.
On San Francisco, McKinnon rejects the "San Francisco is dead" narrative despite migration hype to Miami and elsewhere. Great companies are still starting in the city and people keep returning. He frames San Francisco as essential infrastructure for tech, with the Golden Gate Bridge as a metaphor for a physical identity gateway connecting different areas of the ecosystem.