How far have AI agents really spread?

What the research shows

Agents, AI systems that plan and carry out multi-step tasks with little human input, are moving from pilots into production, but mostly in large companies. In McKinsey's 2026 survey, 40% of respondents from organisations with revenue above $1 billion said they are scaling AI agents in at least one function, up from 27% a year earlier. Among smaller organisations the share was unchanged at 22%.

Use is concentrated in a few functions. Across all respondents, IT, knowledge management and software engineering lead, with 10% to 12% at the scaling phase. AI high performers are more than three times as likely as others to be scaling agents in most functions: 32% in knowledge management and 30% in software engineering, against 10% and 9% of other respondents.

Coding agents are the fastest-moving category. About two in ten organisations are scaling software coding agents, rising to 31% of large enterprises, and 32% of respondents say their organisation decided against buying at least one software product or feature because it could build it with agentic coding tools.

Capability is improving quickly. Stanford's 2026 AI Index reports that AI agents' success on the OSWorld computer-use benchmark rose from about 12% to around 66% in a year, although agents still fail roughly one task in three.

Governance is becoming the constraint. Gartner forecasts that over 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. It lists AI agent identity among its top strategic technology trends for 2027 and expects a quarter of enterprise security breaches to be linked to AI agent abuse by 2028.

What comes next

• The gap in agent adoption between large and smaller organisations is likely to widen over the next year.
• Agent identity, permissions and audit trails are becoming core security requirements.
• Coding agents will move more software from buy to build, changing how technology budgets are spent.

What it means for leaders

• Start where workflows are well defined and value is measurable, such as IT and knowledge management.
• Give every agent an identity, least-privilege access and a named human owner.
• Track the cost of each completed task, not just usage.
• Revisit build-or-buy decisions as coding agents mature.