The Signal · Column 046

An AI agent will run your hiring bar, not fix it

Andrei van Noordt · 3 September 2026 · 2 min read

Phenom is spending this autumn on the road, inviting talent teams to test its AI agents for hiring, development and retention at the season's biggest HR events. That is the shape of every conference floor I hear about this year. The module is out and the agent is in. Something that sources, screens, schedules, nudges people through learning paths and flags who might be about to leave, all of it running live on a stand, all of it visibly faster than the team doing it by hand today.

The reading most people miss is what is actually being demonstrated. Every one of those agents automates coordination. Not one of them decides what good looks like. That definition arrives from the buyer, already written or already vague, and the agent inherits it on day one. If your screening criteria are really a proxy for a school and a logo, you have not bought better judgment. You have bought the ability to apply weak judgment to ten times as many candidates, behind a clean interface and an audit trail that makes it look deliberate.

You cannot automate a judgment you have never written down.

This is where the repricing happens. When every competitor runs comparable agents on comparable data, the speed of your process stops being an advantage anyone will pay for. What stays scarce is the person who can say precisely what a strong hire looks like in your context, defend that standard to a hiring manager, and check whether the machine is reproducing it. That is not the coordinator layer and it is not the sourcing layer, both of which these tools are aimed squarely at. It is a smaller number of senior talent people with real judgment, sitting closer to the business, accountable for outcomes rather than throughput.

For hiring leaders, the practical read is simple. Before you book the demo, write down the standard, the actual criteria that separated your best hires from your worst ones over the last two years. Then make the vendor run their agent against it. If the agent cannot reproduce the decisions you already know were right, you are not looking at a hiring system. You are looking at a faster way to be wrong.