Companies are adding AI workers faster than governance tools

New research suggests agentic AI is moving from experiments to day-to-day work, with active users of OpenAI’s Codex up more than fivefold in the first half of 2026. DoubleU.ai says enterprises need governance systems now, as many still cannot clearly define who AI agents act for, what they can access, or how to shut them off.

Why it matters: - AI agents are starting to function like employees inside companies, with access to data, business systems and day-long assignments. - Many organizations are adding that capability faster than they are building controls for identity, permissions, approvals and auditability. - DoubleU.ai says the gap creates operational, security and accountability risks as agentic AI scales.

What happened: - DoubleU.ai, Inc., the company behind DBLU, warned that businesses need to treat AI-agent management as an organizational discipline, not just a software deployment issue. - The warning follows new research, The Shift to Agentic AI: Evidence from Codex, which found active users of OpenAI’s Codex increased more than fivefold during the first half of 2026. - The study found that more than 10% of users managed three or more concurrent agents during a typical week. - Since the start of 2026, the share of users assigning at least one task estimated to require more than eight hours of experienced human work increased nearly tenfold.

The details: - Adam Harriss, founder and CEO of DoubleU, said companies are beginning to manage AI workers before they have built a management system for AI. - Harriss said every meaningful AI agent should have a job description, identity, manager, defined purpose, authority level, approval limits and a record of what it did. - DoubleU says many businesses still cannot answer basic questions about AI agents, including who the agent is acting for, what job it has, what data it can access and who is accountable if it fails. - IBM Institute for Business Value research with Oxford Economics found 87% of surveyed executives say they have clear AI governance frameworks, but fewer than 25% have fully implemented and continuously review tools to manage bias, transparency and security risks. - DigiCert research found 90% of organizations have discussed AI governance at the executive or board level, but only half have formal AI governance programs. - The same DigiCert research found only 53% can fully trace AI outputs back to the underlying models and source data. - DBLU is building a Trust Layer designed to work at runtime and manage AI identity, purpose, permissions, delegated authority, human approvals, governed memory and auditability. - DoubleU says enterprise software was built around the assumption that a human was behind every important action, and AI agents break that assumption.

Between the lines: - The shift is not just about more AI use; it is about delegation, which raises the stakes for governance. - The research and DoubleU’s pitch point to a coming management problem: companies may need processes for non-human workers that resemble human HR, permissions and compliance systems. - That suggests AI adoption is moving from task assistance toward autonomous execution, even as oversight lags.

What's next: - Harriss is available for interviews on AI workers, unmanaged agent risk and the need for a new management model for non-human labor. - DoubleU is positioning DBLU as infrastructure for enterprises that want to govern AI agents at runtime instead of after the fact. - The broader test will be whether companies can scale agentic AI without creating new security, legal and operational blind spots.

The bottom line: - Companies are deploying AI agents like workers, but many still lack the governance systems to manage them like workers.