AI Fluency Sounds Like a Skill. Right Now, It's Barely a Guess.
The UK government wants 10 million workers trained in AI skills by 2030. It's a genuinely ambitious target, backed by free courses and a cross-government unit tracking AI's effect on the labour market. The direction is right. But a recent piece from Wouter Durville, CEO of assessment platform TestGorilla, makes an argument worth sitting with: training more people is the easy part. Knowing whether it actually worked is the part nobody's solved.
Ten definitions, ten different answers
Durville's point, drawing on TestGorilla's own 2026 survey of roughly 2,000 senior hiring leaders across the US and UK, is that "AI fluency" has become one of those phrases everyone uses and nobody agrees on. Ask ten hiring managers what it actually means and you'll get ten different answers. The survey found 37% of organisations set the bar at simple tool awareness, knowing an AI tool exists, not being able to use it well, adapt it, or judge when it shouldn't be used at all. That's a low ceiling for something increasingly treated as a hiring requirement.
One in five leave it to gut feel entirely
The more striking finding: one in five organisations leave AI-related assessment entirely to whichever hiring manager happens to be in the room that day. No shared rubric, no consistent criteria, no baseline. Two identically capable candidates could get two different outcomes purely based on who interviewed them. Durville calls this the default fallback for any skill category that's new and moving fast: organisations lean on instinct because the infrastructure to do anything else doesn't exist yet.
Training completion isn't capability
The sharpest distinction in the piece is between completing a course and demonstrating a capability. The first is trivial to count, certificates issued, hours logged. The second is genuinely hard to verify, and conflating the two produces upskilling programmes that report strong completion numbers while nobody can actually confirm the underlying capability improved. Given how fast "AI skills" is now appearing in job postings across almost every sector, that gap matters more with each passing month, not less.
What actually closes the gap
The uncomfortable implication is that supply-side training, more courses, more certificates, doesn't solve anything on its own if the demand side, employers actually hiring and assessing people, still has no reliable way to tell who's genuinely capable versus who's just aware a tool exists. Durville's call is for shared frameworks and real benchmarks. That's the right instinct, but it points at a gap that's fillable now, not just something to wait on policy for.
Assessing genuine AI judgement, not just tool awareness, means watching how someone actually works with AI on a real task: where they used it, what they changed, what they rejected, and why. That's the exact shape of the Reflection feature already built into PRODICTA's candidate assessments, a structured, honest account of how someone approached a task, including their use of AI, rather than a self-reported checkbox or a hiring manager's gut call.
The government's training target is a genuinely good start. The measurement problem underneath it, telling capability apart from awareness, is the harder piece, and it's the one worth solving now rather than waiting for policy to catch up.
Sources: this piece references analysis and survey data originally published by Wouter Durville, CEO of TestGorilla, in theHRDIRECTOR.