AI agents are rapidly appearing in enterprise stacks, but most remain isolated in use cases rather than integrated into core workflows. While 90.3% of companies report using AI agents, only 23.3% have them in production and just 6.3% have fully integrated AI into their marketing stack.
Adoption is high because AI is easy to deploy in isolated tasks. Integration lags behind because integrating these results into workflows governed by a system of record is much more complex. In martech, the real constraint is not access to AI: it is aligning probabilistic outcomes with deterministic systems without breaking control, compliance or consistency.
The data shows that organizations do not replace SaaS with AI. They layer probabilistic AI onto the deterministic SaaS systems that still run the business. The challenge is to make these systems work together without creating fragmentation or loss of control. THE agentic stack provides this model, and it varies considerably depending on the size of the company.
Deterministic SaaS and probabilistic AI play different roles, but must operate in the same stack. Registration systems remain the foundation. They store data, apply rules and answer a question: what is true?
AI agents interpret situations and decide what actions to take. They answer a different question: what should happen next?
In its simplest form, the agent stack works like this.
- Context = safeguards: Price rules, product availability, legal and brand rules define what is authorized.
- Intent = situation: What the customer wants and what they are trying to do defines what happens.
- Agents = decision: Reconcile the two to decide what to do
It enables AI to run on SaaS. Integration becomes more critical, but also more complex to control, as decisions now depend on the orchestration of data, rules and context across multiple systems in real time.
Your customers are searching everywhere. Make sure your brand introduces himself.
The SEO toolkit you know, plus the AI visibility data you need.
Start free trial
Start with
How the agent stack works in practice
Here is a simple example. A customer asks for the price of a product via chat.
In a traditional stack, this triggers a search. The system retrieves a price based on predefined rules. The answer is correct, but is not relevant to the customer.
In an agent stack, the same request becomes a coordinated decision. The agent retrieves pricing rules, product constraints, and contractual agreements from systems of record while also evaluating customer context such as behavior, timing, channel, and profile.
- Customer context defines who the response is addressed to. It reflects the current situation of the client, not just its stored attributes.
- Content context defines what can be said. This includes pricing logic, product availability, brand tone, and regional or legal boundaries.
The agent combines the two, crafting a response that aligns with the company’s rules and the customer’s moment. The result is precise and relevant. The right price becomes the right message, delivered in the right way.
How the agent stack changes by company size
The agentic stack evolves based on changes in how intelligence is defined, integrated, and controlled, not by adding more tools or agents.
Small businesses and scaleups are often the most aggressive adopters of martech and AI. They rely on tools to drive growth, which is reflected in both higher relative spending on martech and their approach to integration.
More than half of SMBs (53.6%) rely on iPaaS solutions such as Zapier, Make or n8n to connect their systems, compared to just 20% in enterprise environments. They are also adopting AI through accessible entry points, with 32.1% integrating agents through iPaaS or automation platforms, compared to just 8% in enterprises. This allows for rapid experimentation, but distributes business logic across tools and workflows.
As complexity increases, the limitations of this approach become visible. Mid-market companies are starting to formalize their stack, combining iPaaS, pre-built integrations, and selective custom work. Decision logic begins to move beyond individual tools and a layer of explicit intent begins to emerge.
In enterprise environments, integration is moving toward control and ownership. Nearly three-quarters (72%) rely on custom integrations, compared to 53.6% of SMBs. Enterprises are also integrating AI more deeply into assistants and core platforms (52% vs. 46.4% in SMBs), while facing much greater challenges. Integration frictions reach 68% (compared to 41.1% in SMEs), governance constraints 48% (compared to 26.8%) and cost observability 44% (compared to 17.9%).
Agentic maturity is defined by how effectively organizations integrate systems and govern decision-making through them. As businesses grow, the challenge shifts from providing information to controlling where and how decisions are made across an increasingly interconnected stack.
Retail as an example
Retail provides a useful example of how the agentic stack evolves as organizations grow. This example also clearly plays out within a single vertical.
Let’s look at two perspectives: overall stack maturity and size and, more specifically, one category: tag integration and management.
Overall maturity increases with company size. Small retailers have an average maturity of 2.6, medium retailers 2.8, and large retailers 2.9. Stack sizes are also increasing, from approximately 60% of retail stacks in small businesses to all stacks in enterprise environments.
Integration tells a different story. This category allows businesses to collect customer data and connect systems, allowing data to flow between platforms, create personalized (AI) workflows, and execute agent-driven decisions across the entire stack.
However, as stacks grow, it becomes more difficult to connect systems, manage data flows, and maintain consistency, widening the gap between capacity and coordination.
Small retailers create tightly connected stacks focused on direct revenue impact. E-commerce, CMS, CRM, customer service and performance marketing tools are often linked via iPaaS solutions. Agents already support use cases like product content generation, ad optimization, and customer interactions. But decision logic remains distributed across tools, making consistency difficult to scale.
Mid-Sized Retailers move towards coordination. As campaign volume increases and more channels are added, systems are integrated in a more deliberate way. Agents begin to operate through workflows and decision logic becomes more explicit.
Large retailers operate at a different scale and build their stack around integrated systems of record, including CDP, CDW, PIM and MRM, supporting large volumes of data and campaigns. Agents coordinate decisions across these systems, from pricing and promotions to personalization. At the same time, increased complexity makes it more difficult to maintain control over decision-making.
In all three cases, the trend is consistent. Not only does the stack get bigger, but it also becomes more difficult to manage. The move is to enable execution to control decisions. This is the real change introduced by the agent stack.
