The UK government wants to give 10 million workers AI skills by 2030. It’s an ambitious target, ably backed by free training courses and a cross-government unit to track AI’s impact on the labor market. The direction is right. But ambition and outcome are not the same thing.
To me, the big concern is this: you can train 10 million people in AI and still not know what you’ve got. Because you don’t have a definition for AI fluency, no established rubric for what the start and end state of AI upskilling would look like, and no programme-industry shared definition for what the companies require.
The definition gap is wide, the measurement gap is even wider
AI fluency has become the hiring world’s favorite phrase. It appears in job descriptions, LinkedIn profiles, and now government policy. But ask ten hiring managers to define it and you’ll get ten different answers. Ask them how they assess it, and the conversation tends to get uncomfortable.
That’s not a knock on hiring managers. It’s a structural problem. AI fluency is a broad concept applied inconsistently across industries, job levels, and organizations. According to our 2026 State of Hiring for AI Fluency report, a survey of approximately 2,000 senior hiring leaders across the US and UK, 37% of organizations set their minimum bar at tool awareness. Knowing that a tool exists. Not being able to use it, adapt it, or make a judgment call about when it shouldn’t be used. Just awareness.
That’s a low ceiling, and it matters. Because if the UK’s upskilling program trains workers to be aware of AI tools, and 37% of the employers they’re applying to consider that sufficient, we may be measuring the wrong thing – on both sides of the hiring conversation. For a software engineer, it might mean the ability to build and deploy models. For a marketing manager, it might mean using AI tools to generate content and analyze performance. For a customer service lead, it might mean knowing when to hand off to automation and when not to. These are not the same skill. They don’t map onto a single training module. And they certainly can’t be assessed with a single checkbox on a job form.
AI-related upskilling in local, individual hands
One in five organizations leave AI-related assessment entirely to individual hiring manager discretion. No shared rubric. No consistent criteria. No baseline for comparison.
That number deserves to sit for a moment. In a fifth of surveyed organizations, whether a candidate’s AI capability is deemed sufficient depends entirely on the instincts of whoever is doing the interview that day. Two candidates with identical skills could receive different outcomes depending on which manager they happened to meet.
This is what we call the subjectivity trap; it’s a default setting. When a skill category is new and fast-moving, organizations tend to fall back on gut feel because the infrastructure to do otherwise doesn’t yet exist. The problem is that gut feel doesn’t scale, doesn’t standardize, and doesn’t survive scrutiny when organizations need to report on workforce capability to boards, investors, or regulators.
Upskilling without measurement is a leap of faith
There’s a meaningful distinction between completing a training course and demonstrating a capability. The first is easy to count. The second is harder to verify. And when we confuse the two, we create programs that are very good at reporting training completion rates and less confident about actual workforce transformation.
According to LinkedIn’s 2025 Work Trends data, AI-related skills mentioned in job postings have increased significantly across virtually every sector, some reports pegging a “one-line” mention in 95% of the job postings. The market is moving fast in one direction – more training, more hiring criteria, more AI fluency requirements, without building the measurement layer that would tell us whether any of it is working.
What the government’s program needs to account for
The UK’s 10 million worker upskilling target is a supply-side intervention. It creates a pool of workers with AI training. That’s genuinely valuable. But supply-side alone doesn’t solve the problem if the demand side, that is, employers can’t accurately identify, assess, and hire for what’s been built. Among UK organizations we surveyed that have not defined AI fluency, 47% say they simply haven’t gotten around to it yet. This isn’t indifference, nor confusion, just a capacity issue.
Governments and industry bodies have a role to play here that goes beyond course delivery. Setting shared frameworks for what AI fluency means at different skill levels and in different sectors would give employers something to benchmark against. Encouraging or requiring organizations above a certain size to demonstrate how they assess AI capability in hiring would shift the conversation from inputs to outcomes. It requires the kind of definitional clarity that other professional skills – accounting, legal, clinical – have developed over decades. AI is moving faster than those fields did. The frameworks need to move faster too.
