The Governance Rollback Warning
The most critical warning for C-suite leaders this week comes from Gartner. Their latest analysis suggests that as many as 40% of enterprises will have to scale back or decommission their AI agents by 2027. The reason is not technological failure but a failure of governance.
Organisations are falling into a binary trap: they are treating AI agents as either completely locked down or fully trusted. These uniform controls are creating massive vulnerabilities.
Overly strict policies are driving employees to use unapproved, shadow AI tools, while miscalculated trust is granting agents access to systems they should not touch.
Gartner’s Senior Director Analyst Shiva Varma notes that because accountability for outcomes remains with the organisation, autonomous agents require the most rigorous governance, including continuous monitoring, enforced guardrails, rapid rollback mechanisms, and clear ownership for agent behaviour.
This triangulates perfectly with a commentary piece published in Fortune this week by a risk and operations leader with nearly three decades of experience. The argument is blunt: waiting for perfect information before building a governance structure is not caution but … abdication.
AI is already happening inside your organisation, widely and quietly, well ahead of any governance structure. Employees are running customer data through consumer tools. Engineers are deploying models no one in legal has reviewed. The question is not whether to govern, but whether anyone is IN CHARGE.
The winning firms are not necessarily the ones with the best models but those who build the governance muscle early enough to deploy AI with confidence, speed, and accountability. The AI Governance Committee is no longer a nice-to-have initiative for next fiscal year. It is an urgent structural necessity.
Why You Cannot Hire Your Way to AI Fluency
If you are relying on your hiring process to solve your AI capability gap, you are likely measuring the wrong things.
A 2026 report from TestGorilla, surveying nearly 2,000 senior hiring leaders across the US and UK, found that 59% of organisations made a bad AI hire in the past year.
These were candidates who spoke confidently about AI in the interview but could not apply it on the job.
The problem is that AI fluency has overtaken experience as the top priority (53% prefer strong AI fluency over deep domain expertise), but the hiring infrastructure has not caught up. 71% of organisations have formally defined AI fluency, but only 50% have built internal criteria to measure it.
Interviews are designed to evaluate communication, not execution. A candidate can spend a weekend learning terms like ‘agentic workflows’ and ‘prompt chaining’ and describe them convincingly without ever having built one. As Jason Miller, head of people intelligence and AI at Natera, noted in the report: “Putting ChatGPT on your resume is the equivalent of saying proficient in Microsoft Office”.
The downstream effects of these bad hires are measurable. When someone can describe AI fluency well but CANNOT apply it effectively, it leads to slower execution, inconsistent output, and misplaced confidence in AI-generated work that nobody on the team is equipped to verify.
To fix this, hiring leaders must stop asking candidates which AI tools they use and start asking them to walk through a workflow they redesigned, what changed, what broke, and what they verified. The best AI hires do not just know the language. They can show you the work.
This piece first appeared in Talent Pools Nexus, our weekly read for leaders in research, insights and consulting. Subscribe on LinkedIn.
