Back to insights

Agentic AI and Media Buying: Building the Foundations for Smarter Automation

BlogNovember 11, 2026
By Louis Naghten, Group Partner - Integrated Display

Working with clients exploring how agentic AI could reshape media buying and, more importantly, what they need to build today to be ready for that future has pushed me beyond the usual transformation conversations and into the operational reality of what change actually requires. 

What could agentic layers in media buying look like in practice? And what needs to change for them to work at scale?  

My background is media buying, not LLMs, coding or deep tech. These projects mean getting hands-on with how agentic systems could work in real-world media organisations. The surprising conclusion? A lot of it comes down to plumbing. 

The Next Media Buyer Won't Buy Media

A familiar narrative is already emerging around agentic AI. AI agents will automate media buying. Campaigns will optimise in real time. Traders will spend less time in spreadsheets. Reporting will become faster. Everyone wins. 

Some of that is likely true. But it may not be the most important shift. 

The real transformation is that media buying is moving from doing to delegating. Competitive advantage won't come from having an agent that presses buttons faster than everyone else. It will come from having the systems, governance and decision frameworks that tell those agents what they can do, what they should optimise for, when they need to stop, and who remains accountable when things don't go to plan. 

The future of media isn't autonomous execution. It's trusted autonomy built on clear rules, quality data and human oversight. 

From Buying Media to Designing the System

For years, innovation in programmatic has largely focused on making existing processes more efficient. 

We automated repetitive tasks. We improved bidding algorithms. We built more sophisticated dashboards. Yet the underlying workflow remained largely unchanged. Humans interpreted the brief, built the plan, set up campaigns, monitored performance and decided what happened next. 

Agentic AI changes that operating model because it has the potential to reason across multiple connected decisions. It can interpret objectives, develop plans, take action, learn from outcomes and adapt its recommendations accordingly. 

That's a meaningful shift. 

We're moving from automating tasks to delegating decisions, within clearly defined boundaries. And boundaries matter. 

Media buying has never been a single decision. It's an interconnected chain of choices. Which audience? Which signals matter? Which supply path should we use? What balance between efficiency, control and brand suitability makes sense? How should budget be allocated? 

Today, those decisions are often connected by people, processes and spreadsheets. Tomorrow, agents may connect more of those threads. 

But only if the foundations are strong enough to support them. 

The Unsexy Stuff Becomes Strategic

The least glamorous parts of media operations could become some of the most valuable. 

Taxonomy. Naming conventions. Signal quality. Permissions. Audit trails. Escalation paths. Brand safety policies. Measurement frameworks. 

It's infrastructure. 

But in an agent-powered environment, infrastructure becomes a growth enabler. 

Humans are remarkably good at working around complexity. We interpret inconsistent briefs. We understand local nuances. We know when two dashboards are telling different stories. 

Agents don't have that flexibility. 

If your operating model is inconsistent, agentic systems won't remove the problem. They'll amplify it. 

That's why organisations should start with a different question. Not: Which agentic platform should we invest in? But: What do we need to fix before an agent can make decisions safely and effectively? 

Three areas stand out:  

1. Taxonomy 

A shared language across campaigns, audiences, products, markets and outcomes. Without consistency, agents can't compare signals or make informed decisions. 

2. Guardrails 

Not high-level policy documents, but machine-readable rules. Budget limits, audience exclusions, supplier preferences, brand suitability thresholds and escalation triggers need to be codified in ways an agent can actively follow. 

3. Signals 

Measurement needs to move closer to the point of decision-making. The more rapidly performance and outcome signals become available, the more effectively agents can learn, optimise and adapt. Feedback loops become operational assets, not post-campaign reporting exercises. 

The organisations that create advantage won't necessarily be the fastest adopters of automation. They'll be the ones that can make a good decision once, encode it clearly, and apply it consistently at scale. 

Accountability Gets More Complicated

Imagine, hypothetically, an agent recommends moving 30% of a budget from one market to another based on performance trends. A human approves the recommendation. The outcome underperforms. Who's accountable? 

The person who approved the decision? The team who designed the governance framework? The stakeholder who defined the optimisation objective? These questions will have legal, organisational and commercial implications. 

Audit trails are important, but they are only part of the answer. 

Organisations also need clarity around ownership. Before any agent goes live, teams need to define: 

  • Which decisions agents can make independently 
  • Which decisions require human approval 
  • Which decisions should trigger immediate escalation 
  • What recovery and reversal processes look like 

Without that clarity, governance becomes reactive rather than strategic. 

The Supply Path Question Few Are Discussing

Another challenge emerges when agents begin interacting with other agents. Imagine your buying agent is instructed to maximise reach within strict efficiency targets. A publisher's agent is designed to maximise yield and inventory value. Those objectives may conflict.  Human negotiations rely on context, judgement and compromise. Agent-to-agent negotiation introduces a new layer of complexity. How much flexibility should each side have? When should humans intervene? How are competing objectives balanced? 

The industry is only beginning to explore these questions, but they are likely to become increasingly important as agent ecosystems mature. 

Measurement Stops Being a Mirror

Today, measurement often functions as a reflection of what happened. We analyse results, identify successes and failures, and apply those learnings to the next campaign. 

In an agentic model, measurement becomes something much more dynamic. It becomes an input to decision-making. 

Measurement provides the signals agents use to optimise, act, learn and escalate. It shifts from retrospective reporting to a real-time feedback mechanism. For marketers, that's a compelling prospect. 

Imagine media investment continuously optimising towards genuine business outcomes and incremental impact, rather than proxy metrics. But achieving that requires confidence in the underlying data. Outcome definitions need to be consistent. Signals need to be trusted. Data latency needs to be understood. Teams need clear rules around when agents can act autonomously and when humans remain in control. 

An agent is only as effective as the feedback loop guiding it. 

Performance and Brand Are Different Challenges

One area often overlooked in discussions about agentic AI is that not every marketing objective behaves the same way. Performance marketing provides relatively clear signals. Conversions, CPA and revenue outcomes create measurable feedback loops that agents can use effectively. Brand marketing is different. 

Objectives such as consideration, preference, relevance or emotional connection are harder to define, slower to measure and less deterministic. That doesn't mean agentic approaches won't play a role. It simply means the applications may look different. 

In performance, agents may make increasingly autonomous decisions within established parameters. 

In brand-building activity, agents are more likely to support planning, setup, optimisation and reporting while humans continue to lead strategic interpretation, creative judgement and audience understanding. 

The opportunity becomes even more powerful when richer signals can be introduced into the workflow. 

Brand lift studies, survey data, attention metrics and other outcome measures can all help create more meaningful optimisation frameworks. 

This is an area already advancing. Integrating these signals directly into media decisioning environments has the potential to make brand outcomes more actionable and measurable in real time. 

Humans Move Upstream 

The question isn't whether agentic AI replaces media buyers. The more interesting question is how the role evolves. 

As agents take on more operational and process-driven activity, human expertise becomes increasingly focused on areas where judgement matters most: 

  • Defining business objectives 
  • Designing measurement frameworks 
  • Establishing governance structures 
  • Evaluating trade-offs 
  • Interpreting cultural and market context 
  • Making strategic decisions

The media leaders of the future may spend less time adjusting bids and more time asking: 

  • Are we measuring the right outcomes? 
  • Which data can we genuinely trust? 
  • What level of autonomy is appropriate? 
  • Where should human oversight sit? 
  • Can we explain every major decision? 
  • Can we reverse it if necessary? 

Those are leadership questions, not platform questions, and they're likely to be far more important. 

The Real Risk

The biggest risk isn't autonomy. It's autonomy without explanation. Governance is often framed as a constraint on innovation. In reality, governance is what enables innovation to scale responsibly. 

The challenge emerges when systems begin making decisions faster than organisations can understand, explain or defend. Budgets shift. Audiences evolve. Supply paths change. Performance improves, but if nobody can explain why, trust quickly erodes. That's why transparency and auditability will be critical. 

Every meaningful action should have a record of: 

  • The signal that triggered it 
  • The rule that was applied 
  • The decision logic used 
  • The permissions involved 
  • Any approvals required 
  • The outcome achieved 

When visibility exists, autonomy becomes manageable. Without it, organisations risk creating systems that are faster, but ultimately harder to trust. 

Start Smaller Than the Hype 

Agentic AI often invites grand predictions about autonomous campaigns, self-optimising media and machine-to-machine negotiations. Some of that future may arrive, but the smartest starting point is usually much more practical. 

Focus on repeatable workflows where the value is clear and the risk is manageable. 

Planning scenarios. Taxonomy creation. Quality assurance. Pacing recommendations. Reporting summaries. Supply-path analysis. Then ask three simple questions: 

  1. Is the value meaningful? 
  2. Is the data ready? 
  3. Is the governance model clear? 

If the answer is yes to all three, test and learn. If not, strengthen the foundations first. The organisations that succeed won't treat agentic AI as a large-scale automation project.  They'll build trust progressively. First, agents support human workflows. Then, agents operate within controlled parameters. Over time, greater autonomy becomes possible because the foundations have earned it. Less dramatic, perhaps. Far more likely to work. 

What Really Matters 

Agentic AI may make parts of media buying faster, but speed alone is not a strategy. The real opportunity is to transform complex, fragmented operating models into clear, scalable systems of decision-making. 

Not to remove human judgement, but to make it more consistent. The next generation of media leaders may not spend their time buying media in the way we recognise today. Instead, they'll design the rules, signals and governance frameworks that allow media investment to perform intelligently on their behalf. 

In an agentic future, that might prove to be the most valuable capability of all. 

Share this article

or