AI is already making more of the day-to-day decisions behind media execution, from monitoring performance to optimizing spend. The temptation is to treat that growing capability as a roadmap toward full autonomy: if AI can eventually do something, the goal should be to remove the human from it.
We think that’s the wrong target.
The best media operating model isn’t fully automated or fully manual. It’s one where AI has clear decision rights based on the risk, reversibility and strategic importance of the decision it’s making. Some tasks should happen autonomously. Some should be automated up to the point of approval. Others should remain firmly human-led. The opportunity isn’t to get humans out of media. It’s to stop requiring them for decisions where they add very little value.
Give AI decisions, not just tasks
Most conversations about AI adoption start with tasks: what can we automate? Reporting is repetitive, so automate reporting. Pacing takes time, so automate pacing. QA is manual, so automate QA.
That can create meaningful efficiency, but it doesn’t fundamentally change how a media team operates. People still sit at the center of every decision, with AI making individual parts of the process faster.
The bigger shift happens when you define what AI is actually authorized to decide.
We think about that in three levels.
- Autonomous: AI detects and acts without waiting for a human, within predetermined guardrails.
- Recommend and approve: AI identifies the action and presents the evidence, but a human authorizes it.
- Human-led: AI can inform the decision, but ownership stays with the person making it.
That distinction matters because not every media decision carries the same consequence. Automatically correcting pacing within an agreed tolerance is very different from restructuring a campaign or moving a significant amount of budget between channels. Treating both as simply “automatable” ignores the thing that should determine autonomy: what happens if the machine gets it wrong.
Automate where the decision is frequent and reversible
The strongest candidates for autonomy are decisions that happen repeatedly, are governed by clear rules and can be corrected quickly.
Anomaly detection is an obvious example. AI can continuously monitor hundreds of signals and surface deviations faster than a person manually checking dashboards. Pacing can work similarly: once acceptable boundaries are established, the system can make small adjustments without requiring someone to approve every movement. Platform automation can also handle bidding, delivery optimization, audience expansion and creative routing when it’s working from strong inputs and clear objectives.
These are exactly the decisions where requiring constant human intervention can make the system worse. A person becomes a bottleneck in a process where speed and volume matter more than judgment.
But autonomy should have boundaries. The system needs to know how far it can move before the nature of the decision changes. A small pacing adjustment within an agreed range might happen automatically. A major budget reallocation shouldn’t necessarily receive the same freedom.
That’s where approval becomes critical.
Put humans at the points of consequence
The middle layer is arguably where the most interesting AI operating model is emerging. AI can do the analytical work, identify the opportunity and recommend an action, while humans retain authority over decisions with greater financial, strategic or reputational consequences.
A system might spot that one channel is outperforming expectations and recommend moving budget toward it. Rather than asking an analyst to find that opportunity manually, AI does the monitoring and analysis continuously. But once the proposed reallocation crosses an agreed threshold, the decision moves to a human.
The same principle can apply to creative rotation, significant investment changes or activity in regulated environments. Brainlabs’ agentic model explicitly separates autonomous activities such as anomaly detection and pacing from recommended actions requiring approval and human-led decisions such as campaign structure changes and new builds. Every agent can therefore operate against a documented level of decision rights rather than receiving blanket permission to act.
The point of human oversight isn’t to double-check everything the machine does. That erases much of the value of automation. It’s to concentrate human judgment at the moments where context, accountability and consequences matter most.
Human ownership should move upstream
As AI absorbs more execution, the role of the media practitioner should move further toward the inputs and decisions that determine whether the system performs well in the first place.
Platform automation is a good example. Increasingly sophisticated algorithms can optimize delivery and targeting, but their performance still depends on the quality of what they’re given: the creative, conversion signals, product feeds, business objectives and constraints. When those inputs are weak, giving the algorithm more autonomy doesn’t fix the underlying problem.
That changes where human expertise creates value. Instead of spending time making individual bid adjustments, practitioners can focus on creative direction, testing strategy, cross-channel coordination, measurement design and deciding which business outcomes the machines should optimize toward. AI can also support those decisions, but it shouldn’t quietly become the owner of the strategy simply because it executes more of it.
In other words, the more execution AI owns, the more important the inputs become.
Don’t measure AI adoption by how much you’ve automated
This is where media teams can easily optimize toward the wrong goal. If success is measured by the percentage of workflows automated or the number of agents deployed, the incentive is always to automate more.
A better measure is whether AI has improved the economics and quality of the media operation. Has it reduced time spent on repeatable execution? Can teams respond to performance signals faster? Are fewer errors reaching campaigns? Are people spending more time on decisions where their expertise changes the outcome?
The answers may lead different organizations to different levels of autonomy. A highly regulated advertiser may deliberately require human approval for decisions another brand automates completely. That’s not lower AI maturity. It’s a different risk threshold.
The end state isn’t a media team where AI owns everything.
It’s one where every decision has the right owner. Machines handle the high-volume decisions they can make quickly and reliably. Humans retain authority where judgment and accountability carry more weight.
The question for media leaders, then, isn’t “What else can we automate?”
It’s “Which decisions still require a human, and why?”
