Before You Hand AI Agents the Keys, Hire the People Who Hold Them

Forty-two per cent of companies expect their AI agents to be making decisions without human approval by 2030. Five per cent have the controls in place to let them.

That's from BCG's Applied AI Index, released on 30 September, which surveyed more than 1,300 CxOs and senior leaders across more than 20 sectors.

Closing the gap between those two numbers is mostly about who you hire.

What the research actually tells you

The line I'd underline is this one. BCG found that companies with all six of its agent controls in place across the business generate three times as much value from agentic AI as companies with just one.

BCG also says 70 per cent of the effort in an AI transformation should go into people, organisation and processes, and only 10 per cent into the algorithms. That matches what I see in briefs. Businesses are quick to hire the people who build agents and much slower to hire the people who own them.

The roles that hold the keys

Controls sound like policy. In practice, every one of them is a job somebody has to do.

An owner for every agent that matters. Ownership is one of the controls BCG names: a named person who can change, pause or stop an agent. That's usually someone who knows the process the agent works in, not the person who coded it. I've written before about the agent manager role, and it's still the one most businesses are missing.

Someone who sets the rules. What an agent may do, which data it can touch and when a human has to sign off. That's the remit of an AI governance specialist, and it needs to be in place before the agents go live, not after the first incident.

Someone who checks the work. Agents drift. Somebody has to test them before launch, keep watching them afterwards and roll them back when something goes wrong. That's a generative AI engineer skill set, and it's worth briefing separately from whoever built the first prototype.

Someone senior who joins it all up. Without one executive owner, every team ends up with its own agents and its own rules. My guide to every role in an AI team shows where each of these fits.

Why this lands differently here

BCG's sample is global, and most of the mid-market Australian businesses I speak with are earlier on the curve. I'd treat that as an advantage. You get to put the controls in before the agents arrive, instead of retrofitting them across a dozen workflows nobody fully mapped.

If you're hiring into this

Hire the owner before the builder. If your next AI hire is another engineer, ask who will be accountable for what that engineer ships.

Plan the roles, not just the agents. BCG found 55 per cent of the most AI-mature companies do strategic workforce planning for an agentic world, compared with 17 per cent of laggards. Deciding which roles should exist is the work.

Borrow if the market is thin. These roles are new, and people who have done them well are scarce. A fractional leader can set up the controls and hand them over to your team. The Bot, Build, Borrow, Buy framework I use with clients helps you decide which way to go.

If I were starting this week, I'd list every agent you have running or planned and write next to each one the name of the person who can pause it. Any agent without a name is where your first hire should go.

If you'd like a hand working out who that should be, get in touch and we can talk it through on a quick call.