Eighteen months is less time than a typical one-year cliff plus onboarding. It's also, looking back, long enough for AI to change what hiring means at a level most 'future of work' predictions from a few years ago undersold. This isn't a forecast. It's a look back at what actually happened: which roles quietly disappeared from job boards, which ones didn't exist 18 months ago and now have their own comp bands, and what the shift means for the next 18 months, which will move faster, not slower, than the last ones did.
What quietly disappeared from job boards
The roles that vanished weren't announced as obsolete, they just stopped getting posted, in clusters that tracked specific tooling maturity points rather than a smooth decline. First-line support roles defined purely by answering common questions thinned out as AI-native support tooling matured. Junior roles whose entire job was a repeatable, well-specified task, first-pass data entry, boilerplate code generation, basic content drafting, shrank the fastest, because those tasks were exactly what AI tooling got reliably good at first. The pattern across all of them: the roles that disappeared were defined by a task, not by judgment, and task-defined roles were always the ones most exposed.
What got created that didn't exist 18 months ago
At the same time, a set of roles went from nonexistent or niche to having their own comp bands and dedicated job postings. Evaluation and AI-quality ownership, treating 'is this good enough' as a full-time job rather than a side task, moved from rare to standard at any company shipping AI features seriously. Human-AI workflow design, figuring out exactly where a human needs to check an AI output and where they don't, became its own specialty rather than something engineers improvised. Forward-deployed and embedded AI talent, brought in to own a specific outcome fast without a full-time hiring cycle, went from a niche arrangement to a recognized, named category companies budget for deliberately.
| Thinned out | Emerged |
|---|---|
| First-line support (task-defined) | AI evaluation / quality owner |
| Junior data entry and basic drafting | Human-AI workflow designer |
| Layer-only coordination management | Forward-deployed / embedded AI engineer |
| Generic 'AI engineer' as a catch-all title | Specific AI ownership roles tied to a real outcome |
The unit of hiring changed from 'a role' to 'an outcome'
Eighteen months ago, most hiring plans were still built around filling a role defined by a function: 'we need a data scientist,' 'we need a support rep.' The shift underway now is toward hiring against an outcome an individual owns end to end, with AI tooling filling in whatever part of the function that outcome doesn't strictly require a human for. This sounds subtle but changes the job posting itself: instead of a list of required skills tied to a function, the posting increasingly describes an outcome ('own onboarding retention end to end') and leaves the exact mix of human judgment and tooling leverage to the hire to figure out. Companies that haven't made this shift are still writing job postings for a hiring model that's already 18 months out of date.
Headcount planning: from growth-first to leverage-first
The clearest strategic change is in how companies plan headcount at all. Eighteen months ago, a growth plan and a headcount plan were nearly the same document, more revenue meant more people, roughly proportionally. The companies now outcompeting slower peers plan headcount around leverage instead: what does one person, with the right tooling, credibly own, and only add a person when that ownership genuinely maxes out, not when growth alone suggests it's time. This is the direct, practical form of the broader thesis that headcount growth is no longer a reliable proxy for output growth, and it shows up first in how these companies write their hiring plans, not in a slogan.
What the next 18 months looks like
The pace of the last 18 months, not a hypothetical future, is the best available baseline for the next 18, and every signal points to it compressing further rather than leveling off. The gap between companies that have made the leverage-first shift and companies still running a growth-first headcount plan is becoming a hiring-speed gap as much as a tooling gap: the former can stand up a new outcome-owner in days using embedded or forward-deployed talent, while the latter is still writing a job description for a role definition that's already behind. The practical takeaway for anyone building a hiring plan today is to build it around outcomes and leverage now, not to wait for a future version of this shift that, on the evidence of the last 18 months, will already have happened by the time it feels urgent.