Over the past 12 months, I have seen the campaign brief, long a neglected part of media planning, become the starting point for nearly every agentic workflow. Upload a brief, and the system turns it into a campaign, audience, or optimization plan with minimal input. Yet, while the brief gives an agent valuable details about the campaign and audience, it cannot provide enough context on its own.
Agentic media does not need a better brief. It needs persistent context.
There’s no question that the brief is valuable, but it is just one tree in a forest of context. Whether it was pre-AI or in today’s integrated media environment, I have never seen a single brief fully capture the client, campaign, and strategy. There’s necessary context, whether that’s specific learnings from the previous campaign regarding a certain platform or a shared history of preferences and nuances in how the brand shows up in market. Without a broader system in place in this era of agentic media, teams and agents keep traversing the same forest, rebuilding well-worn paths instead of preserving what they have learned and using that information to chart the way forward.
Fragmented Workflows Rebuild the Same Complexity
The lumascapes that act as a map to our industry show us how individually rational point solutions can create a collective tax on the media dollar. Just take one glance at this year’s iteration, and you can see how the original programmatic ecosystem took more than 15 years and enormous capital to develop.
However, in this day and age, agentic workflows need neither; anyone who can describe a workflow can build one. Yet the risk with these ‘vibe-coded’ solutions is that organizations, while well-intentioned, are recreating the same complexities and costs that characterize the broader media landscape, just inside their own walls, and in a matter of months.
Right now, we see the same pattern being repeated across all manner of organizations, as every new agentic workflow rebuilds context from scratch.
There has to be a better solution.
We have spent years reconciling data across disparate systems, identities, and taxonomies. And now, it seems we are about to spend more time reconciling our beliefs across the broader industry.
So let’s refocus on the actual problem of applying AI to seemingly every workflow. It’s not just about what’s gained, like execution speed or efficiencies, but what’s also lost: the erosion of context due to fragmented workflows.
When the accumulated knowledge, experience, convictions, and choices within an organization are lost as agents take over more and more work, decisions lose the reasoning behind them.
The one thing that’s missing, and is even a gap in ‘human-in-the-loop’ agentic workflows, is the context that informs what a person would instinctively do to solve a problem or further develop a solution or idea. Think about all the disparate Slack messages, meetings, and emails that are exchanged between coworkers on a daily basis. All of that decision-making that ultimately informs the decision your team makes. With more and agentic workflows rolling out across businesses, agents may be saving us time, but we’re missing that persistent context that guides the work.
"The brief is valuable, but it is just one tree in a forest of context… I have never seen a single brief fully capture the client, campaign, and strategy."
For media buying specifically, context accumulates in places a brief rarely captures. It lives in secondary objectives, buried references, the content a CFO distrusts, and the signals that define success. These insights often emerge from unreported findings, campaign optimizations, and relationships between team members built over time and shared experiences. Excluding them from agentic workflows may not cause an obvious failure, but it certainly limits what those workflows can achieve. When data or systems disagree, the industry has specifications, contracts, and discrepancy procedures. But human choices, motivations, and reasoning have no equivalent. An agent may make the best decision available from the context it has stored in the system, but that context is often incomplete.
Beliefs become visible only when they produce a positive or negative result, and each new decision builds on earlier agentic choices rather than the underlying rationale. Agentic workflows should have access not only to the brief, but also to what has been tried, what changed on the platform, how those changes performed, and what the brand and its media partners have learned together.
Persistent Context Must Become Infrastructure
To be clear, I’m not suggesting that we simply add an extension for our agents to send us a Slack message every time the AI may need clarification. Everyone already has plenty of interruptions.
Instead, we need contextualized data that accumulates over time and is available to agents at runtime. Governed knowledge environments, organized as tenancies for clients, sub-brands, initiatives, and campaigns, can develop and protect that context. Making it available across agents in real time creates the kind of persistence a knowledge-driven system requires. We can also build maintenance into the system, avoiding the unfortunate fate of a client wiki or knowledge base left to wither and die on the vine, or, more likely, under a pile of unorganized documents in a team’s Atlassian instance or Google Drive.
At PMG, we are addressing part of this challenge through Alli, our proprietary operating system. Ask Alli works across tools using a shared, governed data foundation, while Alli Business Insights allows teams to incorporate brand-specific context into research that informs planning and creative development. These capabilities provide a foundation for making relevant knowledge available within the workflow. In Alli Buyer Cloud, agents benefit from tenancy controlled context around media, decisioning, performance and persona and deep client and agency knowledge bases. Preserving the reasoning behind a decision remains an essential part of the work, particularly when that reasoning depends on experience that has never been formally documented.
Once, American abstract painter Kenneth Noland said, “For me, context is the key—from that comes the understanding of everything.” His wisdom is applicable far beyond the art world. For us, when the systems we build understand only one aspect of the work, the output is often one-dimensional. Think of a project management system that only includes an abstract name and the deadline for an assigned task. We all know there’s a lot more information needed to shepherd a project across the finish line successfully. For agentic systems specifically, persistent context transforms the practical, step-by-step framework of a workflow into an output that can be trusted.
Better Decisions—not More Agents—Will Define Value
As we build, we should ask how context persists, how it helps us build on what we have learned, and how it keeps us from repeating the mistakes of the last campaign. None of us are being judged by the number of agents we deploy, but by the quality of their work. Look beyond time saved and workflows optimized. Ask whether downstream decisions remain aligned with upstream intent.
Three steps to improve context in your next workflow
Marketing teams can begin with one recurring workflow, such as weekly budget allocation or campaign planning, and examine the information people routinely supply when reviewing its recommendations.
1. Document the reasoning that sits outside the brief. Review the last campaign with the people who managed it and record the decisions a new team member would struggle to explain. Include why a channel was excluded, which tests changed the plan, and where business constraints took precedence over reported performance. Keep those explanations alongside the evidence that informed them.
2. Give that context an owner and a review date. Maintain an approved reference for the workflow, with clear sources and someone responsible for updating it. Review it when objectives change, or new results challenge an earlier assumption. Where your tools support it, make this reference available to the agent each time the workflow runs.
3. Test whether the context changes the recommendation. Run the same planning question with the brief alone and then with the additional context. Have an experienced practitioner assess whether the second recommendation respects known constraints and incorporates relevant learning. Record the corrections it still needs so the team can identify what information is still missing.
In the end, persistent context is what turns faster execution into better decision-making.
Every company will have to justify the value, cost, and outcomes of their agentic workflows. As access to intelligent models becomes commoditized and barriers to entry fall, the technology under the hood will matter less and less. The ability to preserve and apply context will increasingly determine whether agents can make decisions faster and compound that advantage over time.
