Marketing teams have spent two years racing to give AI agents more access: more of the customer data warehouse, more autonomy over segmentation, more permission to act without a human checking the output first. Ataccama’s decision to open source a data quality layer for an emerging AI agent standard is a quiet admission that the industry built the access before it built the trust, and now has to retrofit it.
The Access Race Skipped a Step
Ataccama this week said it will open source a converter that plugs governed business definitions and data quality signals into Apache Ossie, an open semantic specification the Apache Software Foundation is now incubating after Snowflake introduced it as Open Semantic Interchange. The pitch is straightforward: an AI agent can already be told what a field like “revenue” means. What it usually cannot do is check whether the revenue number it is reading right now is complete, current or safe to act on. Ataccama’s converter is meant to close that gap by attaching a live trust signal, not just a static definition, to the data an agent touches.
“AI agents need that same level of operational perspective,” said Jessica Smith, Chief Product Officer at Ataccama. “An agent may know what revenue means, but not whether the revenue data it’s reading is complete or current.” That is not a hypothetical concern inside marketing organizations that have spent the past year handing agentic tools control over segmentation, budget allocation and message sequencing. It is a description of exactly what those systems have been missing.
The Counter Argument: This Is Solved Elsewhere, and Slower Is Not the Answer
The obvious objection is that marketing does not have time for another semantic layer. Every AI vendor pitch of the past eighteen months has been about compressing the distance between a marketer’s intent and an agent’s action, and a data trust checkpoint sounds like exactly the kind of governance step that slows that down. Speed is genuinely the competitive variable right now: the team that gets its agentic segmentation live first captures the response-rate lift first.
That objection gets the tradeoff backwards. A quality check that runs in milliseconds before an agent acts is not what is slowing marketing down. What actually slows a marketing organization down is discovering, three campaigns later, that an agent has been segmenting against a customer field that has been stale for six weeks, and then spending a quarter rebuilding trust in the whole system internally. The cost of skipping the check does not disappear, it moves downstream and compounds. Josh Klahr, Head of Product Management at Snowflake, made a version of this point about portability rather than speed: “Ataccama is extending that portability to data quality, giving organizations a consistent way to make those signals available across platforms rather than rebuilding them for each environment.” A signal that travels with the data, instead of living in one vendor’s walled garden, is precisely what makes a quality check cheap enough to run everywhere instead of expensive enough to skip.
What It Means for the Marketing Leader
This is the same governance question raised by the shift toward exception based rather than approval based control of agentic marketing systems: once a human is no longer reviewing every action before it happens, the quality of the data feeding that action becomes the entire safety net. Salesforce’s own push into agents that run segmentation and activation with minimal human sign off only works if the underlying customer data is trustworthy enough to act on without that sign off. Vendors are not adding data quality layers because it is a compliance checkbox. They are adding them because the marketing organizations buying agentic tools are starting to ask what happens when the agent is confidently wrong.
A marketing operations leader evaluating any agentic tool right now should be asking the vendor a version of Ataccama’s own question: what does this system do when the data it needs is incomplete, contradictory or simply old, and does it tell anyone before it acts, or after. If the honest answer is “it acts anyway,” that is the product roadmap gap to press on, not the autonomy feature to celebrate.
The Call
Marketing has treated data quality as a hygiene project for years, something the analytics team cleans up on a quarterly cadence. Agentic execution turns that cadence into a liability, because an agent making decisions every hour cannot wait for a quarterly cleanup. The vendors building agent standards are already redesigning around that reality. Marketing leaders who keep treating data quality as background maintenance, rather than as the actual control layer on their new autonomous systems, are the ones who will discover the gap the expensive way.
Source: Ataccama
