AI Isn't Replacing Your Martech Stack. It's Being Layered On Top of It.

AI isn't replacing your martech stack. It's being layered on top of it. That's the starting point. And it changes everything about how you should think about your stack architecture.

Frans Riemersma, founder of MartechTribe and co-creator of the MartechMap with Scott Brinker, joined The Oddinary to walk through a model he calls the Agentic Stack. He built it to answer one of the most important architectural questions in marketing technology right now: how do you actually combine SaaS and AI?

The core tension is straightforward. SaaS is deterministic. If-then-else, rule-based. A lead gets qualified, it goes into a bucket. You don't want AI improvising on that decision. But AI is probabilistic. Put the same prompt in ten times and you'll get eight similar answers and two surprises. That's a feature, not a bug. If you use it in the right places.

The layers

Frans breaks the Agentic Stack into two main sections.

The bottom half is deterministic: your hyperscalers (cloud, storage, LLMs from big tech), your bought martech stack (CRM, marketing automation, CDP), and your built software (product configurators, customer portals, price calculators). This is your foundation. It's not going anywhere.

The top half is the AI layer, and it has three levels of its own. First, an intent model layer where you train LLMs according to your company's rules, compliance, and brand guidelines. This is built, not bought. Next, bought AI agents, subscription-based and mostly out-of-the-box. And finally, custom-built AI agents that sit on top of everything else.

"Your stack is a mess. And it's probably close to perfect."

The pattern repeats throughout: buy, build, buy, build. It's never either/or. Nothing is out-of-the-box enterprise-wide anymore. It's always a combination.

Systems of record versus systems of context

Systems of record hold your core customer data, PII, brand assets, and messaging. Deterministic. Stable. These are your foundation. Systems of context sit in the AI layer and answer a different question: how should we act right now?

They pull from the deterministic data but use it probabilistically, in real time, contextually, relevantly. The key ingredient is intent. What is this customer trying to solve, right here, right now?

Frans' advice on integration is refreshingly simple.

"Don't change what already works. Only optimize it when you can, when you need. Don't replace anything if you don't need to."

From monologue to dialogue

Here's the part that stuck with me. Frans argues that AI is actually pushing brands to become more human. In the deterministic world, brands could send messages, build automation flows, and broadcast. One direction. But now customers can go to ChatGPT, Claude, or Perplexity and ask: which product is better? How does this compare? Is this worth the price?

That's a massive power shift. Brands can no longer afford to just tell. They have to explain, listen, and respond in context.

"We go from a monologue into a dialogue situation. And I think that's groundbreaking."

The two essential building blocks for that dialogue? Customer data (knowing who you're talking to) and brand data (knowing what you're saying). Everything else, every agent, every workflow, builds from those two repositories.

Who gets it right first?

You might assume enterprise companies have the advantage. They have the IT teams, the budgets, the scale. But Frans sees something different in the data. Smaller companies are more agile. They experiment faster. They don't need a massive system of record to start doing agentic things.

Meanwhile, 90% of companies say they use AI agents. Only 23% have them in production. Many are completely disconnected from the stack. The outperformers? They're the ones disciplined enough to say: we're experimenting, but we haven't figured it out yet, so we won't scale it prematurely.

The less mature companies externalize. They believe the tool will solve everything. Some even say "AI is our strategy." Frans pushes back hard on that.

"AI is not a strategy. Your customer is your strategy."

The outperformers focus on five use cases that drive 80% of revenue and make those razor-sharp. Then AI suddenly makes a ton of sense.

The bottom line

The Agentic Stack isn't about connecting everything to everything. That's a recipe for a mess. It's about being intentional: knowing which layers to invest in, what to buy, what to build, and where to integrate purposefully.

And maybe accepting that your stack will always be a little messy. As Frans puts it: your stack is a mess, and it's probably close to perfect.

Find Frans on LinkedIn: linkedin.com/in/fransriemersma