top of page

How should professional bodies develop and recognise AI fluency in their members?

  • Writer: Rob Kay
    Rob Kay
  • Apr 17
  • 2 min read

Updated: 5 days ago


Illustration of Software Engineers working with Ai but not having experience for judgment

The FT ran a sharp analysis this week on what's happening to software engineers under agentic AI. 



Job vacancies are up. But only for seniors. Entry-level roles remain flat and the pay gap between top and bottom appears to be widening.



The conclusion many people are drawing is that the future will belong to those who can delegate and direct, not those who can code. The ability to critique and critically review over technical depth.



I understand the conclusion but I think it ignores a difficult reality - the ability to critique and critically review is the result of technical depth - not a substitute for it.



AI enables experienced workers, to leverage their considerable skills at greater speed and scale, because they have their 10,000 hours of technical development behind them.



You don’t get experienced developers, if the work inexperienced ones do is replaced by AI.



One could argue that’s OK because it’s a new skill set that’s required, but if we retrain an entire generation to direct AI rather than do the work who's left with the expertise to know when the AI gets it wrong?



The expertise to catch those errors doesn't come from learning to delegate. It comes from years of doing the work yourself.



AI makes confident errors. It doesn't flag its own blind spots. 



Professional bodies have historically been the custodians for the standard of expertise required. The question they need to be asking right now isn't just "how do we develop and recognise AI fluency in our members?" It's "how do we ensure our members retain the depth to know when AI fluency isn't enough?"



That's a credentialing question. And it's one that frameworks haven't caught up with yet.



And in high-stakes domains like accounting, engineering, law, medicine the cost of an undetected error is serious. It's a compromised audit, a failed structure, a missed diagnosis.



Interested to find models that address this issue. Anything out there?



Comments


bottom of page