Agentic AI in Organisations

• October 2, 2026
• admin

This OECD paper examines how agentic AI is moving from experimentation into practical use, drawing on interviews with 25 organisations across different sectors and regions. Applications include enterprise productivity, customer support, legal work, software development, cybersecurity, supply-chain management and scientific research. Organisations are adopting these systems to improve efficiency, reduce costs, address workforce constraints and create new opportunities, although their levels of deployment vary considerably.

The findings suggest that agentic AI is currently most suitable for structured tasks where outcomes can be verified and risks are manageable. Organisations generally introduce autonomy gradually, retaining human approval for consequential decisions. Preparing employees for these changes is also important, with growing interest in helping staff use and build with AI regardless of their technical background. While these systems may support skills development and provide more actionable insights, excessive reliance on AI could weaken expertise, critical thinking and opportunities for learning through experience.

Major challenges remain, including unpredictable behaviour, cybersecurity vulnerabilities, data governance, accountability and difficulties evaluating system performance. Organisations are responding through human oversight, restricted permissions, testing, monitoring and other technical safeguards. The paper concludes that governance must evolve alongside the technology, with stronger evaluation methods and clearer accountability as systems become more autonomous.