Do the Work Before You Govern the Machine
AI can write your first draft, but only someone who has done the work can tell when it is wrong.
Sign up for my Substack daily AI newsletter here.
See my AI Training course portfolio for corporate Business Leaders here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
You are twenty-five, a few months into your first professional role, and the memo that once cost a junior three late nights now takes you five minutes. Your manager is pleased, and you look productive. Beneath that efficiency, however, a harder question is forming: if the machine is doing the work that used to teach people how to think, when will you learn? Most career advice tells you to stop doing the work and start supervising the machine. The evidence suggests that advice has it backwards.
CONTEXT AND BACKGROUND
The first rung of the professional ladder is under pressure everywhere, and in South Africa that pressure lands on a labour market that was already punishing. In the first quarter of 2026, unemployment among fifteen- to twenty-four-year-olds stood at sixty point nine per cent, and one education leader warned that AI is automating the very tasks through which young workers have traditionally gained experience. At the same time, university recruitment teams report that employers no longer stop at asking what a graduate studied; they want to know what that graduate can actually do.
Those two facts collide. The work that once proved what you could do- the reconciliation, the first draft and the research summary- is precisely the work AI now does fastest. A law firm chief executive and a Stanford researcher argue that this tedious work was never only about output, because it was the training itself. They warn that a person who merely reviews AI output may learn to spot obvious errors without ever building the judgement to produce the work.
INSIGHT AND ANALYSIS
While South Africa’s entry-level market makes the stakes unusually high, the trap facing young professionals is global. The clearest evidence comes from a study of two hundred and forty-four junior consultants at Boston Consulting Group. About twenty-seven per cent handed their task almost entirely to the machine and gained no new skills. A smaller group who kept control of the thinking and used AI for narrow, specific questions produced the most accurate recommendations and continued to build expertise.
This is the illusion of competence. Polished output makes you look capable today while quietly starving the judgement your career will depend on tomorrow. Nobody notices the missing learning until years later, when judgement has become the job.
I have previously written about this when examining a Nature editorial which argued that the real danger is not AI itself but passive use, and that a machine’s answer should be treated as a claim requiring evidence rather than a conclusion requiring acceptance. That principle applies as much to your first job as to your final year of study.
Cognitive governance, the idea that people must remain net contributors to AI-assisted work, translates into three habits for your first years. First, never forward anything you cannot defend line by line. Second, attempt the task yourself before you prompt, then compare your version with the machine’s and study the gap. Third, treat every output as the work of a confident colleague who is sometimes wrong, and find its weakest assumption before anyone senior does.
IMPLICATIONS
A graduating Canadian law student captured the dilemma well. He argued that his generation should use AI heavily while still doing the unassisted analysis through which judgement is built, because the apprenticeship that once guaranteed that analysis is being hollowed out. In South Africa, the stakes are higher still. A non-profit cited by Business Day found more than forty thousand junior digital vacancies unfilled and argued that the barrier is workplace experience rather than qualifications. If AI removes the experience, you will have to build it deliberately.
In practice, that means keeping a running record of where the machine was wrong and what you missed. It means asking your manager to review your reasoning rather than only your output. It also means asking prospective employers how juniors learn when AI writes the first draft. An organisation that cannot answer that question has no plan for your development.
CLOSING TAKEAWAY
You cannot govern what you have never done. The machine will make you fast, but only the work will make you good. Do enough of it, by hand and with intent, to earn the right to judge what the machine produces. In five years, the people who rise will not be the quickest forwarders in the room. They will be the ones who can tell, at a glance, when the machine is wrong.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments