The replacement framing is mostly wrong
Businesses running AI teammates well rarely reduce headcount. What changes is the composition of the work: the routine, documented portion shrinks and the judgment portion grows as a share of everyone’s day.
Volume stops being the constraint
When throughput on routine work is no longer the bottleneck, hiring for speed and capacity matters less. The scarce skill becomes handling the cases that escalate — which are, by construction, the hard ones.
Your team stops doing the easy half. Hire for the half that’s left.
The junior path genuinely changes
Traditionally people learned by doing the simple work first. If that work is automated, you need a deliberate answer for how someone becomes experienced — reviewing the system’s output is one, and it doubles as necessary sampling.
Someone has to own the knowledge
A role that barely existed before: keeping written rules current, reviewing behaviour, deciding scope. In small companies it is part of an existing job, and it must be somebody’s explicitly or it does not happen.
Screen for judgment and for writing
The ability to explain a rule clearly in writing has become directly operational, because written rules are what the system runs on. That is a genuinely new criterion for roles that never required it.
Be honest with candidates
People will ask what is automated and what is expected of them. A clear answer attracts the people who want the judgment work — and evasiveness produces the same distrust as hiding it from your existing team.