British site reliability engineers (SREs) are holding back from letting AI agents loose on production systems, with a new survey showing that trust remains the single biggest obstacle to adoption. The study, conducted by The Register and NeuBird AI in April 2026 among 696 experts, found that nearly three-quarters of organisations have not deployed AI for IT operations (AIOps) at all, while only 8% have it running in live environments.
The findings underscore a deep-seated caution among UK tech teams. When asked what was preventing wider use, 60% of respondents pointed to a lack of trust — far outstripping concerns about return on investment, security and data quality, which each registered around 12–13%. Francois Martel, Field CTO at NeuBird AI, told The Register that the pattern is familiar: 'There's a lot of interest, but not a lot of action.' He noted that while coding agents and content generation tools have taken off because they keep a human in the loop, operations work is fundamentally harder because it runs inside live environments on unseen data with customer-facing consequences.
The trust gap is particularly acute for UK businesses that are subject to the Information Commissioner's Office (ICO) oversight and, increasingly, the EU AI Act's risk-based framework. Martel argues that general-purpose AI agents are a poor fit for SRE problems: 'There are specialised agents that can do a much better job and address some of the concerns around safety, security, guardrails and hallucinations.' His firm's platform is designed to bridge the trust deficit by recording every reasoning step so engineers can audit decisions the way they would interrogate a colleague's incident report.
For UK businesses, the implications are significant. The reluctance to deploy AI agents in production means many firms are missing out on potential efficiency gains in incident response and observability. Yet rushing in without proper safeguards could expose companies to regulatory penalties and reputational damage if agents make faulty decisions in live systems. Martel emphasises that trust is built, not declared: 'Working with AI is a trust-building exercise, and AI has to learn in order to gain trust.'
The survey also revealed that 19% of organisations are currently piloting AIOps, suggesting a cautious but growing appetite. However, Martel warns that teams must choose the right tool category and integrate agents into existing workflows, rather than bolting on a faster responder to a noisy alert queue. Fixing observability at the source — generating high-signal alerts through agentic instrumentation — is key to winning over sceptical SRE teams.