Adobe bets on AI agent orchestration: what changes for devs

During Adobe Summit, Adobe unveiled CX Enterprise, an enterprise platform built to coordinate AI agents across marketing, content, and customer engagement workflows. At first glance, it looks like just another enterprise tool. However, when we look through a developer's lens, the proposal is quite different: it's a layer that connects agents, corporate data, and external applications through reusable skills and integration endpoints.

In other words, Adobe is building a technical foundation for something many teams already try to solve the hard way: getting AI agents to talk to each other, with security and governance.

Moreover, the move comes at a critical moment. According to Adobe's own research with 3,000 executives, data fragmentation and misalignment between tools are the biggest obstacles to scaling AI. Therefore, it's not an exaggeration to say orchestration has become the new bottleneck.

The real problem: isolated agents don't solve enterprise workflows (Adobe)

First, it's worth understanding the context. Many development teams have already tried creating agents with LLMs for point tasks: generating copy, summarizing tickets, classifying leads. However, when the goal is to automate an end-to-end flow, things quickly fall apart.

After all, a real workflow involves:

• Fetching data across multiple systems

• Applying business rules and compliance

• Generating brand-aligned content

• Triggering actions across different tools

• Logging everything for auditing

Consequently, without an orchestration layer, each agent becomes an island. And islands don't scale.

How Adobe structures CX Enterprise under the hood

The platform combines three main blocks: AI agents, reusable agent skills, and Model Context Protocol (MCP) endpoints. All of this sits under a governance and intelligence layer.

For developers, the interesting part lies precisely in the reusable skills and MCP endpoints. In other words, these are what let you define and manage workflows between systems without rewriting integration from scratch on every project.

Furthermore, CX Enterprise relies on Adobe's Experience Platform, which, according to the company, supports more than 1 trillion customer experiences per year. This data layer feeds agents with real context, rather than prompts bloated with irrelevant information.

Two systems that make the magic happen

Adobe introduced two central components:

Brand Intelligence System: captures feedback, rejected assets, and annotations. It then applies that context to content generation.

Engagement Intelligence System: works as a decision engine. It determines actions based on customer lifetime value and pre-defined business goals.

This way, multiple agents can share data and decision logic across different workflows. This matters because it avoids the classic problem of agents making contradictory decisions at different stages of the funnel.

CX Enterprise Coworker: the layer that sequences tasks between agents

Now we get to the part that matters most for those building multi-agent systems. CX Enterprise Coworker is a coordination layer that operates on top of the intelligence systems and translates business goals into sequences of actions.

In practice, it handles:

• Audience segmentation

• Content generation

• Campaign monitoring

• Human oversight and auditability

So instead of you coding all the logic for “which agent calls which agent, and in what order,” Coworker resolves that sequence based on declarative goals. Adobe demonstrated an example where the system collects audience data, produces assets, and tracks performance, all as a coordinated workflow.

For devs used to orchestrated pipelines, the analogy is clear: think of it as an Airflow, but for AI agents with built-in enterprise governance.

Integrations that avoid vendor lock-in (at least in theory)

One point worth noting: CX Enterprise integrates with AWS, Google Cloud, Microsoft, IBM, Nvidia, OpenAI, and Anthropic. It also connects to environments such as Microsoft 365 Copilot and ChatGPT Enterprise.

As a result, Adobe's agents can access corporate data and generate output within these applications. This matters because 75% of organizations, according to data cited by the company, point to data quality and integration as central challenges to scaling AI.

In other words, the bet is on interoperability. However, healthy skepticism is warranted: in practice, deep integrations tend to create their own forms of dependency.

What the survey numbers reveal (and what they hide)

Some figures deserve a critical reading:

• Only 43% of executives consider their data adequate for AI

• 78% expect agents to handle half of support interactions within 18 months

• Only 16% of organizations have already deployed active AI for support

• 70% of consumers expect AI interactions to feel human

Therefore, there's a huge gap between expectation and operational reality. In short, most companies want autonomous agents, but don't have their data foundation ready. That's where platforms like CX Enterprise try to position themselves.

GenStudio also got agents: impact on the content workflow

Besides CX Enterprise, Adobe's GenStudio platform also received updates. It now features AI agents that automate campaign planning, content production, and workflow management.

A Brand Intelligence System integrated into GenStudio adjusts outputs based on feedback and approval cycles. This way, output improves over time, instead of relying solely on initial prompt engineering.

What devs should watch for from here on with Adobe

Regardless of whether you use Adobe products or not, this move signals relevant trends:

First, the MCP standard is gaining traction as an integration layer between agents and systems. It's worth studying.

Second, “reusable agent skill” should become a common concept in enterprise architectures, similar to what microservices were in the 2010s.

Third, governance and auditability are no longer optional. Any serious multi-agent architecture will need to answer “who did what, when, and why.”

Finally, the data showing AI system traffic on US retail sites grew 269% in one year shows that conversational search is already changing user behavior. So, backends need to be ready to respond to agents, not just humans.

Conclusion: orchestration is the next battleground

Adobe didn't invent agent orchestration. However, by packaging governance, integration, and coordination into an enterprise platform, the company helps consolidate an architecture pattern that had already been under discussion in technical communities.

For developers, the message is direct: mastering MCP, multi-agent architectures, and governance patterns will stop being a differentiator and become a requirement. Meanwhile, it's worth watching how CX Enterprise performs off the Summit stage, after all, conference announcements and real-world production tend to be different universes.

Check out our Instagram by clicking here and stay in the loop!