AI Agents as Workers, with Sharon Hunneybell

Most organisations still treat AI agents as software. Their staff do not. Agents are being used as colleagues, confidants and co-developers; they are being handed access to multiple systems, and a good number of them belong to proof-of-concept projects that quietly ended without anyone shutting off the credentials.

Recorded at AdelaideSEC, David Savva-Willett speaks with Sharon Hunneybell, VP of Products at FirstWave, ahead of her talk on a governance framework for AI agents as workers. Sharon describes realising midway through writing that framework that she had missed a step, and that discovery has to come before onboarding, because these things arrive in the workplace unannounced and increasingly as features inside software that is already approved.

The conversation covers where accountability sits when an agent acts, why access should be task-based and time-limited rather than granted to one broad agent, how far traditional HR frameworks actually translate, and what Sharon expects from regulation over the coming months. It closes on her practical checklist: give every agent its own identity, log what it does, scope its access to a single task, review it on a schedule, and know how to offboard it before you build it.