Eighteen months into what was supposed to be the first great automation wave, the balance sheet is more interesting than the headline. The mass-displacement predictions from 2024 did not materialise the way pundits expected. Neither did the universal productivity windfall the other camp promised. What we’re actually seeing, quarter by quarter in the client work, is something the public discourse hasn’t quite caught up to: a quiet, uneven re-shaping of what roles contain.
Our read is that the jobs most affected haven’t been the ones most threatened in the 2024 thinkpieces. Junior white-collar roles — the ones the automation fears were loudest about — have shifted in composition rather than disappeared. Meanwhile, roles almost nobody was predicting would change (senior operations, middle-management coordination) have been quietly transformed, because the capability that matured fastest was coordinated multi-step agentic work, not generative writing.
Five patterns shaping hiring and workforce plans for the rest of the year:
- Junior tasks compress, but junior roles don’t vanish. The first two years of a graduate career used to be mostly repetitive learning work. That work compresses. The roles survive because the apprenticeship function — building judgement — still needs humans practising on consequential problems. Teams that eliminated their junior layer in 2024 are quietly rehiring it this year.
- Middle management gets the biggest shake-up. Coordination, status-tracking and report-assembly — the actual daily contents of a mid-level role in most enterprises — is the sweet spot for agentic systems. The surviving layer has moved from producing coordination artefacts to reviewing and correcting them. The headcount curve has bent, modestly.
- Specialist roles deepen rather than disappear. Tax accountants, compliance officers, clinical coders — the roles everyone assumed would be the first to go have gained leverage, not lost it. One specialist plus a capable system now does the work of several. Organisations that thought they could remove the specialist entirely have almost universally reversed course.
- A new discipline is emerging: AI operations leadership. Someone has to own the evaluation suite, the incident response, the vendor relationships, the cost curve and the roadmap of what to automate next. This role didn’t exist on org charts two years ago; it’s being created under every title you can imagine (director of AI, head of automation, platform lead). Budget for it deliberately.
- The skill premium is on judgement, not on prompting. The hiring conversations we’ve been part of for the last six months have barely mentioned prompting skills. They’ve been about judgement — what to automate, what to supervise, when to escalate. That shift matches what the work has actually become.
The workforce question for 2026 is not “will AI take jobs.” It’s “which humans in which roles are now leveraged by AI, and which are not, and why.” The answer is company-specific, it moves quarter-over-quarter, and it’s much more interesting than the discourse is making it look. The organisations thinking about it concretely — role by role, team by team — are the ones where the numbers are working. The ones still debating the headline are, unsurprisingly, the ones still stuck in pilots.
