Postman survey finds agent API access outpacing identity controls

src-22de100f

Organizations are putting AI agents into production faster than they are defining how those agents should authenticate, according to a Postman survey of nearly 9,000 developers, architects and executives. The findings point to a widening gap between agent deployment and the identity, oversight and retirement practices needed to control access to application programming interfaces (APIs).

A quarter of respondents said their organizations already have agents calling APIs in production. Yet only 13.2% said their APIs are ready for agents to use. The figures highlight a practical challenge for enterprise security teams: an agent may be able to invoke services and retrieve data, while the organization lacks a distinct identity or a reliable way to determine what that agent is permitted to do.

Shared credentials obscure agent activity

Postman found that about four in ten respondents had not worked out how credentials and identity should work for agents. Three in ten said agents use the same credentials as other API clients. When an agent shares credentials with scripts or other clients, its requests may be harder to distinguish, and revoking access specifically for that agent can be difficult.

The leading concern reported by respondents was an agent accessing data or API endpoints it was not authorized to use. A separate identity for each agent can help make access attributable and allow teams to manage or disable an agent without treating it as an indistinguishable client. Postman’s first recommendation is to give every agent its own identity; about a third of surveyed organizations have already done so.

Identity is only one part of the control picture. A third of organizations reported having no formal process for managing agents. Fewer than a quarter maintain an inventory of the agents they run or monitor agent activity, while 9% said they have a process for retiring an agent. Together, those results suggest that some organizations may have limited visibility into which agents are active, who is responsible for them, and how access should be withdrawn when they are no longer needed.

API-first organizations still report credential sharing

Postman compared organizations based on whether they describe themselves as API-first, meaning they design APIs before the software built around them. The API-first group was more likely to inventory agents, assign an owner to each, and provide agents with separate identities. However, it also reported running more agents, and 43.2% said their agents use ordinary shared credentials. The corresponding figure among organizations that are not API-first was 23.0%.

Postman cautioned that these results show a correlation, not that an API-first approach causes credential sharing or other outcomes. The API-first respondents also leaned more toward executives and directors, whose perspective may differ from that of teams handling day-to-day implementation. The comparison therefore should not be read as a definitive measure of operational maturity.

For security leaders, the figures underscore that API design practices do not replace agent-specific access controls. As deployments expand, teams need to know which agents exist, who owns each one, what credentials it uses, and how its access can be reviewed and revoked. Those controls provide a foundation for investigating unexpected API activity and limiting the impact of an agent that acts outside its intended scope.

Human-oriented APIs create another readiness gap

Many APIs were designed for human developers who can consult documentation and ask colleagues for clarification. Agents cannot resolve ambiguity in the same way. Postman noted that outdated documentation was already the top complaint among surveyed developers, a limitation that can also obstruct an agent attempting to use an API reliably.

Respondents identified AI-powered attacks and rogue agents misusing APIs as joint leading concerns for the next two years, with concern about attacks rising compared with the previous year. For enterprise teams, the survey supports a review of agent access as part of existing API governance: establish a separate identity for each agent, maintain an inventory and accountable owner, monitor activity, and define a process to retire agents and revoke credentials. Organizations also need to assess whether API documentation is current and clear enough for the agents they permit to use it.

The survey does not establish that every production agent is poorly controlled, nor does it quantify incidents. It does show that deployment is already underway while identity and lifecycle practices remain uneven. Closing that gap will be central to making agent access observable, attributable and removable.