There is a moment in almost every executive conversation about AI agents when the real question surfaces. It is rarely asked directly. It arrives dressed as a spreadsheet: projected efficiency gains, full-time-equivalent savings, payback periods. Underneath it is a fork with two paths.

Path one: agents as a cost lever. Automate the work, reduce the headcount, bank the saving. The business case is easy to write and the payback is legible to a board.

Path two: agents as a capability lever. Give your existing people agents, keep the headcount, and do things your competitors cannot — faster turnaround, deeper analysis, service levels that were previously unaffordable.

Both paths are real. We think leaders owe their organisations more honesty about the choice than the spreadsheet provides, so here is ours.

The cost path quietly eats its own foundations

The awkward truth about agent deployments is that they depend on precisely the knowledge that cost-path restructuring removes. Agents must be taught how work actually happens — not the documented process, the real one, with its exceptions and workarounds and unwritten judgement calls. That knowledge lives in experienced staff. Trim them first and you discover, mid-deployment, that nobody remaining can explain the exceptions the agent keeps mishandling — and nobody is left to notice when it goes subtly wrong.

There is also a slower cost. Cost-path organisations tell their remaining staff, unmistakably, that helping the agents improve is helping to write their own redundancy. Rational people respond by withholding exactly the cooperation the deployment needs. We have watched this dynamic sink technically sound projects.

The capability path is not charity — it is compounding

The case for path two is not sentimental. An experienced employee with a well-deployed agent is a different economic unit: same salary, materially more output, and — this is the compounding part — their daily corrections and feedback make the agent better in a way no vendor update can. Organisations on the capability path build a private asset: agents tuned by their best people to their actual work. Organisations on the cost path buy the same generic capability as everyone else, minus the people who could have differentiated it.

None of this means headcount never changes. Roles will genuinely transform, and some tasks will stop existing. The distinction is whether displacement is the strategy or an occasional, openly handled consequence of a strategy that is fundamentally about doing more.

Your staff already know which path you are on

Perhaps the most practically useful thing we can report from our assessment interviews: employees read leadership's intent long before any announcement. They read it in what gets measured, in whether training accompanies the tooling, in whether the first pilot lands in a team as an assistant or as a shadow. And their reading determines their disposition — enthusiastic co-designers on one reading, quiet resisters on the other.

That disposition is measurable, and it is one of the strongest readiness signals we know. Which is the polite way of saying: if you have chosen the cost path but are announcing the capability path, your readiness problem is not your data, your platform, or your policy. It is that nobody believes you — and they are right.