Agentic AI Is Fast Approaching. Here’s How to Get It Right

I’m as skeptical about the latest “shiny object” announcement as anyone else. And just coming back from Cannes Lions, companies both large and small had major news to share; the coming of agentic artificial intelligence was the talk of the town.

Right now, we’ve got marketing teams running 80 to 120 different tools and platforms. Half the time, they can’t even talk to each other easily—they are siloed. APIs come to the rescue and bring the necessary platform-to-platform connectivity needed to power today’s adtech ecosystem.

But APIs don’t solve for cross-platform orchestration. If you want to run a simple audience activation campaign across multiple channels, suddenly you need a team of engineers just to make it happen.

So, with everyone on the Croisette racing toward AI agents that promise to automate complex workflows, how can businesses actually start AI-enabling their tech stacks? 

That hope is where the Model Context Protocol comes in. MCP isn’t going to inspire screaming headlines about epochal change for your career. There won’t be any conference sessions about MCPs replacing your job as an award-winning creative. But the potential is very real that MCP is emerging as the standard that will make our agentic marketing future a reality.

The foundation of connectivity

APIs are the OGs of connectivity. They’ve been around since the 1950s (really).

APIs are used to connect today’s tech stacks, but they clearly weren’t built with AI agents in mind. Consider a typical campaign workflow: You want to build an audience in your customer data platform, push it to The Trade Desk and Meta, get reporting back, optimize based on performance, track conversions, and send a summary to your client. Sounds simple enough. But in reality, you’re dealing with different rate limits, varying data structures, and about six different places where the whole thing can fall apart.

Now, imagine trying to get an AI agent to handle that workflow. With APIs, you’d need to teach the agent every platform’s specific requirements, every error condition, every little gotcha that comes with each integration. That doesn’t make business sense, especially when you are talking about 80-120 platforms in your tech stack.

MCP flips this completely. Instead of agents needing to understand the technical complexity of every platform, they communicate through the protocol servers that provide context (the “C” in MCP) about what each system can do and how to work with it. Think of it as giving your agent a really good interpreter for every platform in your stack.

What agentic orchestration actually looks like

The promise here isn’t just about simplifying integrations. It’s about finally enabling the kind of automated workflows we’ve been talking about for years.

I could write into a chat interface right now, “Take our high-value customer segment from last quarter’s campaign. Run it on social and programmatic channels. Aim for a return on ad spend of more than four times, and send me updates every week on how it’s performing.” An agent equipped with the right MCP connections could theoretically execute that entire workflow without me touching another platform.

Again, this is not five years in the future. The major LLM providers have already adopted MCP as the standard: OpenAI, Gemini, and Claude are all supporting it. The future is now.

Why this time should be different

What I like about MCP compared to every other “revolutionary” integration solution we’ve seen is the modularity. You’re not locked into some vendor’s ecosystem or forced to rebuild everything to make it AI-enabled or when you add a new platform.

Managing API-based connections at scale takes real resources. MCP adds a new connection layer that makes those same APIs usable by agents across multiple platforms, extending the value of APIs, not replacing them. 

Think about it from a practical standpoint. If you’re running an enterprise tech stack, you can’t wait for every vendor to build native AI integrations. But you can use MCP as the bridge between your existing APIs and these new agentic capabilities.

The guardrails we actually need

You might very well be wondering whether we’re going to have AI agents running loose with our media budgets. Look, I get it. The idea of autonomous systems making real-time decisions with client budgets should make you nervous.

It comes down to safeguards. MCP works through existing API frameworks, so they inherit the guardrails you’ve already got in place. Plus, you can layer on additional controls. User profiles also limit what agents can do. Mandatory approval checkpoints are an essential safeguard for big decisions, and there are audit trails for everything.

The goal isn’t to eliminate human oversight. The role of professionals’ judgment is the point. Instead of manually moving files between systems and babysitting integrations, you should be setting strategic parameters and making high-level optimization calls while agents handle the operational tasks.

Start small. Maybe it’s just automating audience file transfers. Then maybe it’s basic reporting aggregation. You crawl, then you walk, then you run.

The choice

What most people aren’t thinking about yet is that MCP isn’t just about connecting your own platforms. It’s about adding a new layer of connectivity that other companies can plug into. The question is whether you’re building that orchestration layer or just hoping someone else builds it for you.

We need to be smarter about this tool. Choose your MCP providers carefully. Start with clear use cases. Build governance frameworks before you need them. And please, for the sake of your sanity, don’t try to automate everything at once.

This is the good thing about these early days. We can approach MCP thoughtfully and create something that actually works. Or … we can rush in blindly and create another expensive mess that requires armies of specialists to maintain.