Postman asked nearly 9,000 developers, architects and executives how they’re handling AI agents, and the answers tell a familiar story. Companies put the new thing to work first and left the question of who holds the keys for later. A quarter of the organizations that answered already have agents calling their APIs in production.
Source: Postman
The keys are where it gets uncomfortable. About four in ten respondents haven’t worked out how credentials and identity should work for agents at all. Three in ten let agents log in with the same credentials as any other API client, so the agent shows up looking like every other script on the network. If it starts reaching for data it shouldn’t, the company has no separate identity to rein in or cut off. That is the scenario respondents worry about most: an agent getting into data or endpoints it was never authorized to touch.
Agents got access before they got rules
Picture an agent as a new hire who skipped orientation. It can read your systems, call your services and act on its own, but at many companies it never got a badge, a manager or an exit interview.
Companies are far more likely to have dedicated security controls for their APIs than for the agents calling them. A third have no formal process for managing agents. Fewer than a quarter keep a list of the agents they run, or monitor what those agents do. And 9% have a process for retiring an agent, which means most of these companies can switch one on with no plan for switching it off.
The best-run shops share keys too
Postman sorts respondents by how API-first they are, meaning whether they design the API before the software around it. The companies that call themselves API-first are better organized almost everywhere you look. They’re more likely to keep an inventory of their agents, name an owner for each one, and give every agent its own identity.
They also run a lot more agents, and 43.2% of them let those agents share ordinary credentials. Among companies that aren’t API-first at all, the figure is 23.0%. More agents in production means more of them using borrowed keys, even at the companies doing the most homework. The researchers are careful to say the survey shows correlation and not cause, and that its API-first group leans toward executives and directors, whose view from the top may not match what their teams do day to day.
Most APIs were built for people
A quarter of organizations run agents in production, but only 13.2% say their APIs are ready for agents to use. Most APIs were designed for a human developer who reads the docs and messages a colleague when something doesn’t make sense. An agent can’t do the second part. Outdated documentation was already the top complaint among the developers surveyed, and for an agent it can mean the end of the road.
Respondents seem to sense where this goes. Asked about the next two years, they ranked AI-powered attacks and rogue agents misusing APIs as joint top worries, with attacks climbing since last year.
Postman’s first suggestion is to give every agent its own identity. About a third of the companies surveyed have already done it.

Strategies: CISSP exam prep with ISC2 experts

