Big companies trust their AI agents, but only 7% would let them run solo, like a great new starter who still doesn't get the company card
By Brett Gavaghan, Founder of Adaptd ·
A G2 study of 3,222 enterprise reviews, published on 30 October, found 77% of users trust their AI agent. Only 7% would give it total autonomy. Most want it working inside set limits.
The short version
- G2 analysed 3,222 reviews of AI agents from companies with more than 1,000 staff and published the results on 30 October 2026.
- 77% of these reviewers trust their agent, but only 7% would give it total autonomy, and 60% want autonomy with guardrails.
- If you're setting up an AI agent, give it clear limits and pick one that explains its decisions before you give it more freedom.
Plenty of businesses now have an AI agent doing real work. Very few are letting it do that work without someone keeping an eye on it.
What did G2 find about AI agents?
G2 found that 77% of enterprise reviewers trust their AI agent 'quite a bit' or 'very much', but only 7% would give it total autonomy. The software review site analysed 3,222 reviews from companies with more than 1,000 employees, covering 231 AI agent products, from January 2025 to 23 October 2026.
Most people want something in between: 60% want autonomy with guardrails, 21% want limited autonomy and 11% want a human to approve every action.
What would make people trust AI agents more?
Knowing what the agent did matters more to these reviewers than raw accuracy: 79% picked a fix about visibility or control. Explanations of how decisions were made came first at 32%, then more guardrail settings at 29%, with better reliability on 21%.
There's a gap between getting it right and showing its working. 92% rated their agent's accuracy 4 or 5 out of 5, but 82% said the same about how well it explains its results.
What should a smaller business take from this?
A smaller business can copy the big ones: let an AI agent work on its own inside clear limits, and keep a person on anything that goes to a customer or costs money. That doesn't need an IT department, only a decision about what the agent can and can't do.
My take: start with one job, such as sorting enquiries or drafting quotes, and choose a tool that shows you why it did what it did. Keep in mind this is review data from G2, which sells services to software vendors, not a controlled study. Too small for guardrails? A one-line rule like 'it drafts, I send' counts.
The agent can have the keys to the filing cabinet. The bank account can wait.
Wondering what this means for your business?
That's exactly what the discovery call is for.