AI agents are running your GTM stack, and a third of ops teams can't say how many

A survey of 157 B2B revenue and marketing operations practitioners, published by GTM software vendor LeanData and reported by MarTech on September 25, finds that 93% of go-to-market teams have already put at least one AI agent to work, while nearly a third of the people running those teams cannot say how many agents are currently acting on their own systems.

The respondents, surveyed in May 2026, are LeanData's own customers — revenue operations and marketing operations professionals who already run the company's routing and orchestration software, so the sample skews toward teams that were early to this kind of tooling, not a cross-section of the whole market. Even inside that relatively advanced group, governance has not kept pace with adoption. Thirty percent of respondents told LeanData they had caught an agent taking an action with no audit trail behind it, and just 8% placed themselves in the survey's top readiness tier, "fully optimized." Seventy-nine percent said they were still standing up their first agent use cases or actively scaling existing ones.

Where the gap shows up in the pipeline

For software marketers specifically, the survey's most concrete findings are about collisions, not abstractions. Twenty-seven percent of respondents said multiple tools or agents had contacted the same prospect independently of each other, and 17% had watched a marketing sequence fire while a sales rep was already mid-conversation with that same account. Those are exactly the failure modes marketing ops teams spent the last decade building lead-routing and suppression rules to prevent — and a new layer of agents, each with its own trigger logic and none of them reading the others' state, is recreating the problem faster than most teams can patch it. A lead-scoring model that double-fires an email is an annoyance; an agent that double-books a demo, or emails a prospect a sales rep just told not to contact them, is the kind of thing a CMO has to explain upward.

MarTech, which first wrote up the findings, is owned by Semrush; its coverage of a vendor-run customer survey should be read with that in mind, same as the underlying numbers should be read as a picture of LeanData's install base rather than the broader market.

The root cause is the data, not the agents

Asked what's actually stalling their AI GTM initiatives, respondents didn't point to the agents themselves. Incomplete or inconsistent CRM records topped the list at 45%, undocumented processes drew blame from 37%, and siloed teams from 32%. Seventy percent said poor data hygiene has already degraded go-to-market execution, and 55% named data quality and AI readiness as their single biggest transformation blocker. That ordering matters: teams are layering autonomous decision-making on top of the same messy contact and account records that already caused reporting headaches before any agent touched them, and an agent will act on bad data faster and more often than a human would have.

"AI models are now both powerful and incredible. Yet, we all know they work only as well as the infrastructure that lies beneath them." — Katy Keim, CEO, LeanData

Who owns this, and what to check this quarter

Governance ownership itself splits three ways: a cross-functional committee holds the reins at 42% of organizations, RevOps owns it alone at 18%, and at 19% nobody owns it at all. Sixty-nine percent are using AI features already embedded in their GTM tools, 62% have built custom applications on top of LLM APIs, and 46% run dedicated agent platforms — Agentforce, Copilot and Gemini Enterprise among the ones named — often more than one at once, with three to four separate agents being the most commonly reported count. What respondents wanted most, cited by 31%, was full traceability: a record of every action any agent takes against any customer or prospect record.

None of this argues for turning agents off. It argues for treating each new one as a system that touches shared prospect and customer data, not a marketing tactic you can switch on inside a single tool's settings panel.

The uncomfortable read for marketing leaders is that agent adoption has outrun the unglamorous infrastructure work — clean CRM data, documented processes, one clear owner — that made the last generation of marketing automation safe to run at scale. The 2026 version of that work is audit trails and agent inventories rather than list hygiene and lead scoring rules, but it is the same job, and skipping it costs the same way: duplicated outreach, prospects contacted by systems that don't talk to each other, and no record of why. Teams that treated data governance as a prerequisite for automation, not an afterthought to it, are the ones this survey found closer to that 8% "fully optimized" tier — everyone else is layering agents onto a foundation they already knew was shaky.