AI agents are no longer a concept deck for programmatic advertising. They are live, making real-time buying decisions, and the industry is scrambling to build the governance structures around them before something goes wrong.
The IAB Tech Lab has launched a Programmatic Governance Council specifically to address this. The timing is not accidental. Several major advertisers and agencies are already running AI agents that autonomously adjust bids, shift budgets between channels and pause underperforming campaigns without human intervention.
KSM, a media agency, describes its agent architecture as a "librarian" model. The agent can access data, run analyses and make recommendations, but it operates within defined boundaries. Spend caps, brand safety filters and escalation triggers are hard-coded. The agent cannot exceed a daily budget threshold without human approval.
Bayer has taken a similar approach, implementing spend caps and category exclusion lists that the agent cannot override. The brand has reportedly seen efficiency improvements of 15 to 20 per cent on campaigns managed by agents, primarily through faster reallocation of budget away from underperforming placements.
Efficiency improvement reported by Bayer on AI agent-managed campaigns
The governance challenge is real though. An AI agent optimising for a cost-per-acquisition target does not inherently understand brand safety, competitive separation or media quality. Those constraints need to be programmed, tested and monitored. The IAB's council is working on standardised frameworks for agent behaviour, including audit trails, decision logging and accountability structures.
For Australian advertisers, this is relevant now, not in two years. The major DSPs are all building agent capabilities into their platforms. Google's DV360, The Trade Desk and Amazon DSP all have agent-adjacent features in market or in beta. If you run programmatic at any scale, your buying platform will offer you an agent option within the next 12 months.
Why it matters
Programmatic advertising has always been semi-automated. AI agents push it toward full automation, which creates genuine efficiency gains but also genuine risks. The advertisers who benefit will be the ones who define clear operating parameters before turning agents on, not after.
What to do about it
Start documenting your programmatic buying rules in a format an agent could follow. What are your hard limits on spend per channel? What brand safety categories are non-negotiable? What performance threshold triggers a pause? If you cannot articulate these rules clearly for a human media buyer, an AI agent will not follow them either. Build the governance framework now while the stakes are low and the technology is still being shaped.
