A marketing stack is rarely designed in one go. It grows one purchase at a time: a tool for email, one for social, one for SEO, one for reports. Each was bought for a good reason. Few of them know what the others did.

AI agents now sit inside many of those tools, and it is often predicted that they will shrink the stack. The survey data on what marketing teams are actually doing shows how that is starting: inside the tools teams already have, one job at a time.

What the survey found

Scott Brinker and Frans Riemersma published Martech for 2026 in December 2025. It is built on their AI & Data in Marketing survey of the people who run martech and marketing operations. The authors note that this group leans more tech-savvy than the average marketing team, so read the figures as a view from the front of the field.

90.3% said they use AI agents somewhere in their martech stack, even if only in experiments or limited use.
Only 23.3% had agents in full production.
85.4% use agents to improve what their existing tools already do, and 42.7% to add something new.
Only 30.1% use agents to replace an existing martech capability.

Replacement is more common in larger companies: 36.0% of enterprise respondents use agents to replace some of what their stack does, against 19.6% of smaller businesses.

So the agents are arriving inside the stack, not instead of it. The authors expect more replacements, with a limit: "We expect there will be more substitutions in the year ahead. But mostly for very specific use cases, not wholesale replatforming." For now, the main job of an agent is to help the tools a team already trusts do more.

Where agents are doing the work

The most common internal uses in the survey:

Content production (68.9%): drafting and personalizing content, with a person reviewing it before it ships.
Audience discovery and segmentation (40.8%): finding patterns in customer data and acting on them.
Competitive analysis (35.9%): reading competitors' sites and turning what changed into something the team can act on.

None of these is flashy, and each one is gathering work that used to eat into a marketer's week. The decision about what to do with the result stays with your team: the plan, the brief and the final call.

The data comes first

The report's most practical finding is about data, not agents. 56.3% of respondents said poor data quality (missing, stale or inconsistent data) was a significant difficulty, and it was the biggest issue teams reported with integrated AI systems. The report uses the now common name for getting the right data to the right agent: context engineering.

An agent working from a messy CRM writes to the wrong people, segments on old fields and reports numbers nobody trusts. Clean records, one definition of a customer and a named owner for each field will do more for your results than the next tool.

Five steps for your team

1. Pause new purchases while you review

Set a fixed window for the review and add nothing until it ends. A stack is easier to see when it stops moving.

2. Review use, not spend

For each tool, ask who uses it, how often, and what decision it feeds. A tool your team opens every day can be worth more than a cheaper one nobody opens.

3. Fix the data before adding agents

Start with the records your agents will read: customer profiles, consent, product data and the content library. Decide who owns each one.

4. Ask what each tool would gain from an agent

In the survey, far more teams use agents to improve a tool than to replace one. For each tool you keep, write down what an agent inside it should do and who on your team approves its work.

5. Choose platforms that share one memory

The real cost of a stack is that nothing remembers what the rest did. Favor a platform where the plan, the content, the sends and the results sit on one record of your brand and your customers, whether it works alongside the tools you keep or takes the work on itself. That is the idea behind martech stack consolidation with one brand memory.

The bottom line

AI agents are not going to empty your stack overnight, and the teams in the survey are not trying to. They use agents to get more from the tools they trust, they fix their data first, and they replace a tool where an agent does a specific job better. Your team decides what stays, what goes and what ships. The stack just starts working together.