Only 11–14% of enterprise agent pilots reach production — the rest stall on identity, audit, and access control
The plumbing is standardising faster than the deployments are shipping. Industry estimates put the share of enterprise agentic-AI pilots that reach production at just 11–14%; the majority stall not on model capability but on the governance layer — agent identity, audit trails, and access control. The bottleneck has moved from 'can it reason' to 'can we let it act'.
The gap is diagnostic. When nearly nine in ten pilots fail to reach production and the common cause is identity, audit, and access rather than capability, the conclusion is that the models are ready and the controls around them are not. An agent that can plan and call tools is only deployable once an organisation can say who authorised it, what it may touch, and what it did.
Agent identity is the specific hard problem. A human user has permissions; an agent acting on that user's behalf needs a derived, scoped identity and an audit trail, or every action it takes is an accountability gap. Until that exists, a security or compliance team is right to block the pilot, and most do — which is what the 11–14% figure is measuring.
The read for the market is that the next wave of agent products will compete on governance, not intelligence. The stateless MCP rewrite makes fleets cheaper to run, but cheap-to-run agents still stall at the same wall until identity and access are solved — which is why the enterprise conversation this quarter is about control planes, not capabilities.
The Agentics — The enterprise MCP guide 2026 → · Sourcetrail — AI agents in 2026: protocol updates, enterprise tools, autonomous payments →