Two Identity Conferences, One Unanswered Question
Two conference halls, two weeks apart. Identity Week America at the Walter E. Washington Convention Center on September 2-3, with more than 3,000 attendees, 250 exhibitors and 250 speakers. CYBR.SEC.CON at the George R. Brown Convention Center in Houston on September 15-16, another 3,000-plus across ten specialized tracks.
Different cities, different audiences. Washington drew the identity specialists: biometrics, decentralized identity, ID wallets, verification. Houston drew the broader security profession: hacking, OT security, executive leadership and threat research.
Both agendas converged on the same subject, and it was not identity governance. It was AI.
What was on the agenda
Identity Week America made agentic AI its flagship theme. The program framed the shift from passive AI copilots to autonomous, task-oriented agents that "query system APIs, execute transactions, and spawn sub-agents with minimal human oversight," and put the governance questions directly: should an autonomous agent inherit a user's permissions or hold credentials of its own? What happens to delegation when an agent requests access on behalf of a person? How does any of it work when the permutations of agents and sub-agents cannot be predicted in advance?
Houston went the same direction by a different route. CYBR.SEC.CON launched a dedicated AI track this year after a record number of AI submissions, with sessions on shadow AI, autonomous agents, MCP servers and AI governance, and AI turned up across most of the other tracks too.
These are the right questions. They are also, in most enterprises, several questions ahead of where the operational reality sits.
What sits underneath the agenda
The sessions were about governing autonomous agents. The problem underneath them is older: applications that were supposed to be onboarded last year.
The shape is familiar across the industry. A program has an IGA platform and a coverage number in the seventies. The remaining applications are the ones with no API, or a vendor that charges extra for one, or a network segment the platform cannot reach. Those applications have been on the roadmap through three planning cycles. The certification campaign goes out against the ones that were onboarded, and the audit finding lands on the ones that were not.
It is rarely anyone's most interesting problem. It is usually their most persistent one.
The gap between the agenda and the estate
The industry's attention has moved to a genuinely important frontier. Autonomous agents will hold credentials, act across systems, and create access relationships nobody explicitly granted. Getting ahead of that is not premature.
But an identity program that cannot currently answer who can reach what, and why for the applications it already has is not positioned to answer it for a population of agents that changes hourly.
The uncomfortable version of the question: could this organization prove, today, who has access to what across its estate? Not across the applications inside the IGA platform. Across the estate.
For most, the answer is no. Not through carelessness. Because the applications where the answer is hardest to produce are precisely the ones that never got integrated, and they never got integrated because integrating them costs more than the program could justify.
Why the gap persists
The failure is not in the platforms. SailPoint, Saviynt, Okta and the rest do what they are built to do once an application is connected to them.
The failure is in how connection happens. Every application arrives as a fresh problem. Someone researches how it stores users, decides what counts as an entitlement, works out how to correlate accounts to identities, builds and tests the connector, and negotiates thirty to forty hours of an application owner's time to do it. Then the next application arrives and the same architectural work begins again from a blank page.
Delivered that way, coverage plateaus exactly where the economics stop justifying the next integration. Which is why so many programs end up with the same shape: strong coverage of the applications that were straightforward, a permanent backlog of the ones that were not, and a coverage percentage that has not moved in two years.
Meanwhile the denominator grows. According to Verizon's 2026 Data Breach Investigations Report, the share of employees who are regular AI users on corporate devices tripled in a year, from 15% to 45%, and 67% of them reach those services through non-corporate accounts. Those tools hold credentials. None of them are in anyone's identity governance platform. The coverage ratio did not fall because governance got worse. It fell because the estate got bigger and nobody recounted.
What would have to be true
For the agentic AI conversation to be actionable rather than aspirational, the deployment problem underneath it has to be solved first.
That means the cost of bringing an application into governance has to fall far enough that the ungoverned tail stops being a permanent line item. It means the architectural work (schema modeling, entitlement design, correlation logic) has to compound across applications instead of restarting at each one. And it means that the same machinery has to work for the systems that resist integration, because those are where the answer to who can reach what is least available and most needed.
Governing autonomous agents is a harder version of a problem the industry has not yet solved in its easy version. The organizations that will handle agent identity well in three years are the ones that fix their deployment layer now, while the population being governed is still made of applications and people.
The question worth taking home
Both conferences were worth the time, and the agenda was not wrong. Autonomous agents are coming to enterprise identity faster than most programs are ready for.
But the question that has to come first is the simpler one. Not how we will govern the agents. Just: can you prove, today, who can reach what?
Where Klyro fits
That question is why Klyro exists.
Klyro is the deployment layer for identity governance. It onboards applications into the IGA platform an organization already runs: research, schema modeling, entitlement design, correlation rules and connector code, validated in a sandbox and signed off by a human architect before production.
The work compounds, so each application makes the next one cheaper. Application owners react to a proposed design on day one instead of piecing it together meeting by meeting. The backlog becomes a schedule.
The agents are coming. The applications are already here. Govern the estate you have, and the agents become the next applications in the queue.
Coverage number stuck for two years? Talk to Klyro's experts at hello@klyro.security