A survey of 1,000 engineers shows AI agents with write permissions jumped from 52% to 89% in a year. Of those, 61% require human approval before acting, but 24% write code independently. Organizations now face urgent pressure to build security and governance systems around autonomous agents.
What level of tool permissions do you give your agents? This surveyed 1,000 engineers. And now here's what it says. Agent teams using agents, 89% are giving them write permissions now, up from 52% a year. Do you know what that means? So imagine you have like a Hermes or something like that. Write permissions in which they can go and actually make changes. They can execute. Write humans with human in the loop, it's like 61%. And then write access independently without human approval, 24%. Read only, cannot make changes. And then we don't use agents internally. So 89%. This is up from 52% last year, which means this: people are going to have to invest more in security, more in governance, More on permissioning. I think that's a very good service to have, Neo.
