Your AI stack has 5 layers. Interoperability can’t be an afterthought.

Enterprises aren’t building their AI environments from scratch. They’re layering new capabilities onto the technology stack that’s been in place for years. Some AI capabilities arrive built into the operating systems and tools employees already use, some come from frontier AI platforms, and others from departments investing in specialized AI for functions such as marketing.

These additions aren’t always coordinated across the enterprise. Each AI tool selection may make sense on its own, but that doesn’t mean the pieces will work well together. A recent study found that large enterprises use an average of 183 AI tools, yet 85% of surveyed C-suite executives say their organizations lack the systems to connect them.

The result can be overlapping investments, complicated integrations, and less flexibility as business needs and AI capabilities evolve.

CIOs are often held accountable for how this technology environment performs, even when they don’t make every decision that shapes it. And as scrutiny of what AI investments deliver grows, they need to demonstrate value today without limiting their options tomorrow.

Interoperability is key to achieving that balance. It connects the different layers of an AI environment, allowing enterprises to choose the best technology for each job and add or replace capabilities as needs evolve.

Unlock greater AI value without the lock-in.

Interoperability changes how enterprises can put AI to work. For instance, a frontier model can contribute to a marketing workflow without having to handle every part of it. It may generate a campaign plan, while specialized marketing AI solutions and existing systems handle the audience management, content operations, activation, and measurement needed to execute it. Each model, solution, and system brings capabilities and expertise the others may not have.

Standards such as Model Context Protocol (MCP) and Agent2Agent (A2A) help make these connections possible. MCP is an open standard, providing a common way for AI applications to access external tools and data, while A2A allows agents to communicate and coordinate across systems, so new capabilities can work with what’s already in place.

These connections also make an AI environment more composable, which is particularly valuable when capabilities evolve faster than technology buying cycles. Enterprises can add or replace capabilities without being locked into a fixed set of tools — and employees can access capabilities from different applications through their preferred interface instead of moving between systems.

Making this work depends on choosing well across five layers of the enterprise AI environment and understanding how those layers connect.

Understand how every layer of your AI environment connects.

Each layer of the AI environment calls for a different kind of decision. Evaluating an enterprise AI agreement is different from deciding how AI fits into employees’ operating environment, which solutions and agents teams need, or where specialized marketing AI belongs. And how well these technologies perform depends on the quality and accessibility of the underlying data foundation.

The goal is to make the right choices at each layer while ensuring the pieces work well together. Here’s what to consider as you work through them.

1. Enterprise-level AI: Understand what you’re building on.

An enterprise agreement with a frontier AI provider can create the impression that the company has everything it needs to develop its own AI solutions. Test that assumption against specific business requirements by confirming which capabilities are ready to use, which require custom development or specialized applications, and which are still planned but not yet available.

The answers determine how much work the enterprise is actually taking on. Then examine the systems those capabilities must connect to. Some enterprises discover multiple instances of the same data platform or several digital asset management systems. Mapping what already exists reveals where new AI can plug in and where integration can reduce additional effort.

Decision point: Determine which needs your AI provider can meet directly and where additional development or integration is needed to make its capabilities work.

2. Operating environment: Align AI with how employees work.

An enterprise’s operating environment is often where employees first encounter AI. Built-in capabilities arrive through familiar workplace tools, sometimes without a separate rollout or decision on their part. Whether that familiarity translates into adoption depends on how closely the standard environment matches employees’ preferred tools and continuing workflows. Turning that alignment into sustained use requires deliberately planned training and learning support, especially before expanding AI across the broader business.

As you scale, compare the standard environment with how teams work, checking for institutional knowledge current tools may not consider. Where workflows differ, consider if employees can access the AI capabilities they need through familiar interfaces rather than having to switch platforms. Interoperability can help make that possible without forcing everyone to use the same tools.

Decision point: Check whether your standard operating environment matches how employees operate across established workflows, and whether you can bring AI capabilities into the tools they prefer.

3. AI assistants and agents: Rationalize based on the work.

At this layer, the options multiply. Companies may build their own interfaces on frontier models, develop custom agents, or use open-source alternatives, while departments adopt different assistants. The question is whether that variety serves distinct business needs or simply creates duplication.

Answering this calls for evaluating solutions in the context of actual workflows. Work with the teams using these tools to assess whether they can reliably complete their intended tasks and hand work to one another. A tool may perform well on its own but add complexity if it cannot exchange information or coordinate actions with the systems around it. Looking at the broader workflow can also reveal where functions overlap.

Decision point: Assess which tools to retain, consolidate, or retire, and how open standards or orchestration could help the remaining capabilities work together.

4. Marketing technology: Match AI to specialized workflows.

Frontier models can generate marketing content and draft plans, which leads some IT leaders to treat marketing as a use case a general-purpose model can cover once it’s connected to company data.

But marketing AI isn’t one capability. AI that improves brand visibility in LLM-driven search serves a different purpose from AI that coordinates a customer journey across websites or applications, eventually ending in a conversion. Each relies on different data, specialized knowledge, and systems that turn insights into action. Evaluating only what a general-purpose model can produce risks overlooking those requirements.

That’s why the first step is to work with marketing stakeholders to examine the full workflow, including what happens before and after an AI task. Establish what your existing systems already cover, where the workflow needs specialized AI, and how those capabilities would work together. If a custom build is considered, assess the development time, integration effort, and ongoing maintenance it would require.

Decision point: Identify which marketing needs call for purpose-built AI and how those capabilities must connect with the broader technology environment.

5. Data foundation: Make data ready for AI.

Underneath these technology layers sits the data they all depend on. Models, agents, and specialized applications need reliable, accessible, and well-managed data to deliver useful results. Even when the technologies connect seamlessly, poor data quality can undermine what they accomplish.

One bank I spoke with had invested in several AI tools without seeing the expected results. It eventually returned to its data foundations, recognizing that better tools alone wouldn't solve the problem.

To avoid discovering these problems after deployment, evaluate data readiness against the workflows AI is expected to support. Start with a priority use case and determine whether its data is accessible, sufficiently contextualized, and governed for the intended purpose. For a deeper look at these foundations, see my earlier blog on how to prepare your data for AI success.

Decision point: Determine how well your data foundation can reliably support the AI capabilities you’re deploying, and which gaps you need to address.

Make interoperability a shared priority.

Choosing well across these layers means making interoperability part of every AI decision from the start. IT and business teams need a shared view of what a new capability will contribute, what the enterprise already has, and how the new piece will connect to the rest.

Business teams know the workflows AI needs to support, and IT can assess the integration work, engineering effort, and ongoing maintenance involved.

Bringing them together early lets teams weigh an investment’s potential value against its demands on budgets, people, and existing systems. It also keeps a choice made for one workflow from closing off options for the next, because each piece is chosen with connection in mind.

As enterprises put connected AI to work across more workflows, they also need to operate these capabilities reliably at scale. Doing that raises another set of questions about governance, operational accountability, and how AI-driven decisions translate into action. I’ll explore these in my next blog.

Let’s talk about what Adobe can do for your business.

Get started

New message from advisor