Article
The new question is not only whether to hire — it is also whether to buy
When an AI agent can perform part of a role, headcount planning turns into a mix decision.
Headcount planning used to be a simple question: do we need another person in this team, yes or no. It is no longer simple.
When a well-built agent can absorb a meaningful part of a role, the question becomes a mix: how much of this work should be done by people, and how much by capability we buy or build.
Roles are not replaced — they are decomposed
Very few roles are wholly automatable, and almost every role contains work that is. Intake, classification, first-pass drafting, reconciliation and status chasing are usually the automatable part.
Once you decompose the role, the planning question changes: which slice goes to an agent, and what does the person do with the time that comes back?
Questions worth asking before the next requisition
- Which parts of this role are repetitive, high volume and rule-bounded?
- What is the cost per case today, and what would it be with an agent handling the first pass?
- Who supervises the agent, reviews exceptions and owns quality?
- What happens to the work if the agent is wrong — and how would we know?
- Are we buying a ready-made agent, building one, or hiring and revisiting in two quarters?
The mistake to avoid
Cutting headcount first and hoping the agent covers the gap creates operational risk and internal resistance in the same move.
The sequence that works is the reverse: automate the slice, prove the quality, then let the role evolve around the capability that is genuinely in production.
Hiring and buying are no longer alternatives on a form. They are two levers on the same capacity plan, and the organisations that model them together get both cheaper and steadier.
This article is the English version of a post originally published in Hebrew on our LinkedIn page.