The Agentic Marketing Garage: What Happens When Agents Work Together?
Most of us have used AI by now. Maybe you have searched with Gemini or Perplexity, leaned on an LLM to draft a document, or started experimenting with Claude’s agentic tools in your workflow. AI is changing how we work — and at Adobe Summit this year, we wanted to push that conversation further.
Not can an AI help me with a task? But: what happens when a team of agents tackles that task together?
That question was the foundation of the Agentic Marketing Garage — a first-of-its-kind event at Adobe Summit in Las Vegas, held on Wednesday morning to a sold-out room of 150 customers.
---
Lifting the Lid on the Black Box
When you ask an LLM a question and get an answer back in a few seconds, it feels almost magical. But what is actually happening underneath? What would you see if you could lift the lid?
That is exactly what the Agentic Marketing Garage was designed to show.
We created five live playground environments, each demonstrating a different agentic use case. These were not scripted demos — the questions came from the audience in real time, generating scenarios we had not pre-planned. Every interaction you saw was genuinely live.
---
How the Playgrounds Worked
Each playground featured five agents working together:
- An expert agent trained on Adobe services (including documentation & product knowledge, as well as product capabilities and services)
- A moderating agent to guide the session and govern the work
- Three customer-controlled agents — configured by attendees
The setup mirrored something familiar: a team of people in a room, each bringing different knowledge, perspectives, and approaches to a shared problem.
Here is how a typical session unfolded:
- The moderating agent posed a question — drawn from the audience
- Each agent produced an initial proposal based solely on its existing knowledge
- The audience selected a direction from the responses
- Agents were then given access to Experience League's MCP server — pulling real documentation and reference material in real time
- Each agent produced a revised proposal incorporating that research
- The agents then debated and critiqued each other's work — openly, conversationally, building on each other's findings
- Four separate answers converged into one collaboratively developed solution
- The agents produced deliverable artifacts: a proposed content calendar, a project plan, suggested KPIs, and a division of work
- Finally, agents rated each other's contributions — and named an MVP
What we observed in those exchanges was striking. Agents would say, in effect: "That was a strong direction — I found an additional document that addresses the gap in your proposal. I am incorporating it into my response." The answer improved. The group got smarter together.
This is, of course, exactly how human teams work.
---
Five Playgrounds, Five Use Cases
We ran this experiment across five stations, representing a connected marketing workflow:
| Playground | Focus |
|---|---|
| Experience League Research | Autonomous multi-solution content supply chain research |
| Creative with Firefly Services | Generating imagery and assets for marketing campaigns |
| AEM Document Authoring | Building real web pages using Adobe component structures and style guides |
| Campaign Flow | Developing full campaign proposals and planning frameworks |
| Halo/Experience Platform Offers | Next best offer modelling and customer engagement follow-up |
Together, these five use cases form a coherent arc: research → creative → web → campaign → customer offer. End-to-end, agentic.
---
What We Learned
A few things stood out.
Agents are extraordinarily fast. Each demo ran for approximately eight minutes — but that was slowed down deliberately so people could follow the interactions. In reality, the agents completed the same work in around ninety seconds. When you consider the scope of what was happening — research, synthesis, debate, artefact creation — ninety seconds is remarkable.
The interaction is the insight. Most people understand that an LLM can answer a question. Fewer have considered what it looks like when multiple agents disagree, refine, and converge on a solution in real time. Seeing that unfold — agents genuinely critiquing each other, incorporating new information, evolving their positions — shifted people's understanding of what is coming.
Knowledge currency matters. One of the most revealing moments was the before-and-after: agents producing an initial answer from training alone, then producing a substantially better answer after accessing live documentation. It illustrated the real problem of enterprise AI at scale — not can it answer? but how do you know it is right, and how do you keep it current?
There is a barrier to entry. Configuring agents is still somewhat of a technical task. It is getting easier, and I certainly believe that while agent config is not like 'installing an app' today, it will be in a short space of time. So today that makes this a tall order for a lot of people. It will get easier.
Your work laptop will fight you. Or more accurately, your Enterprise IT policies will fight you. And to some extent, rightly so. The potential of breaches in security and the exposure of data is very high, as organizations wrangle with what policies, procedures and infrastructure needs securing. This makes it more tricky for professionals to experiment with autonomous agents - they have to do this on their own time. This is one of the reasons you can't buy an Apple Mac Mini for love nor money currently.
---
The Bigger Picture
Experience League already receives more than five times as much agentic traffic as human traffic. Research is the obvious first use case — and agents are doing it well. But this event was about what comes next for marketers.
In the not-too-distant future, you will not just have an AI tool that helps you with tasks. You will have a set of agents you orchestrate — agents that work for you, with you, with each other, with Adobe's agents, and with agents from other platforms — to accomplish things at a scale and depth that has not previously been possible.
In many ways, you will become the CEO of your own domain: the expert who sets the direction, while agents handle the research, the execution, the iteration, and the improvement.
The Agentic Marketing Garage was a first experiment in understanding what that looks like. We ran thirty-plus demos across five pods in two hours, with 150 customers watching it happen live.
We learned a great deal together. So did the agents.
And that, as it turns out, is exactly the point.
---
Have questions about agentic AI and Experience League? Join the conversation, ask us a question or give us your thoughts.