Gartner predicts 40% of enterprises will demote or decommission autonomous AI agents by 2027 due to governance gaps that often surface only after incidents. At Snowflake Summit in San Francisco, three digital leaders offered a practical counterweight—framework
By the time an autonomous AI agent is already in production, the mistakes can look less like theory and more like damage.
That timing is exactly what Gartner is pointing to with its prediction: 40% of enterprises will demote or decommission autonomous AI agents by 2027 because governance gaps are often only identified after incidents occur when those agents are running in real environments.
Against that backdrop. three digital leaders shared what it looks like to push agents into production without treating them like magic. The lessons were delivered at the Snowflake Summit in San Francisco. where Matt Luizzi. VP of analytics at wearable technology specialist Whoop. Madeleine Want. VP of data at sports specialist Fanatics. and Sriram Sitaraman. CIO at software specialist Synopsys. each described how they’ve tried to turn agent hype into measurable work.
Gartner’s forecast lands on a painful reality for teams betting their workflows on AI: when governance isn’t nailed down early, failures tend to be discovered late—after the agent has already been trusted.
In that sense, the most consistent thread across the summit talks wasn’t “more AI.” It was more structure.
Luizzi’s team at Whoop built their agent rollout around frameworks that could be repeated and evaluated. Whoop collects biometric data 24/7 to power its health and wellness insights, with Snowflake supporting the firm’s internal analytics services. In that system. Luizzi said agents play an increasingly important role. particularly Snowflake CoCo. the technology specialist’s coding agent for developers and data engineers.
“We’ve been using CoCo for several months now, and started with just the analytics team,” Luizzi said, describing the initial focus as people who could quickly look at a query response and say whether it was correct or not—then work to scale that process.
He said the organization moved to more formalized evaluation frameworks and started rolling agents out at scale. For Luizzi, the point wasn’t just speed; it was making experimentation repeatable. He said software engineers at Whoop deploy A/B tests and use CoCo to analyze the results. propose the next feature. test it. and iterate.
“This approach is rapidly accelerating the way that we’re shipping not only business value, by automating the experimentation framework, but also the customer value,” Luizzi said.
He also tied progress to preparation already done on the data side. Luizzi said Whoop was fortunate that the underlying plumbing for agentic explorations was already in place because the company’s data is centralized on the Snowflake platform. The team used Snowflake’s Cortex AI service to start testing agents and learning lessons. One of those lessons. he said. was that “context was everything. ” which meant leaning into the semantic layer and making sure the context is in a structured place.
Want, at Fanatics, described how her organization learned that governance and data quality weren’t just prerequisites—they were what made agents actually answer questions effectively.
Madeleine Want manages data engineering. data science. and machine learning across Fanatics’ betting and gaming division. with this work supported by the Snowflake platform. She said that during early experimentation, the team wasn’t sure what would stick. What did stick. she said. was tightly linked to the condition of the underlying data and the governance of it—because the better those were. the more easily the LLM could derive meaning and answer questions.
“It certainly wasn’t the case 18 months ago. ” Want said. pointing to a shift from earlier experience building bespoke machine learning models. She said it was hard to believe that importing a third-party model and putting it on top of the data could work for analysis. but that approach is now embedded in how Fanatics does things.
Want also traced the move from exploration to exploitation. She said success came early in domains that were well bounded in context, and where expert analysts understood the business domain top to bottom and could coach the agent.
Over time, she said the organization’s results improved. The investment needed in the context layer decreased. as did the degree of supervision an agent required before it could start answering questions autonomously. She said scaled evaluation frameworks helped the team increase confidence in how agents were answering when they weren’t being looked at.
“Our ability to measure the accuracy of the answers is increasing,” Want said, adding that the frameworks are helping with confidence in agent outputs when the goal is effectively to let them run.
As those successes accumulated, Want said the scope of agents expanded beyond analytics. She described a shift in demand from other professionals who see positives and want to explore agents, not just use them.
Fanatics. Want said. still uses Snowflake’s interfaces and agents. but it is embedding APIs and responses into other third-party tools so people can do more with data-powered insights. “Users want to go further and do more with operational use cases,” she said. “People are demanding to be able to access those insights through a variety of different channels and consumption mediums. because they need to be able to use data where they’re working.”.
Sitaraman’s account at Synopsys brought the conversation to value—specifically, how teams are trying to monetize data instead of treating AI agents as a standalone experiment.
Synopsys is a long-time Snowflake customer. Sitaraman said. using the data platform and its agentic services such as CoCo to power decision-making processes. About 18 months ago. he said the company recognized that AI agents could fulfill tasks typically handled by junior employees—running quick queries. creating graphs. and deriving insights.
“We took advantage of that capability, and we said, ‘OK, look, if we create a knowledge agent, we can start deploying it in multiple dimensions,’” Sitaraman said.
He offered examples including a revenue agent for the finance department that runs reports and a debug agent for the ticketing system associated with the firm’s data centers.
Before committing. Sitaraman said the team assessed AI across three dimensions: the quality of results. time to results. and cost of results. He said the team discovered AI had a positive impact in all three areas. calling it “a significant breakthrough.” The implication. in his telling. was that earlier tradeoffs weren’t required the way teams had expected.
“In the past, you had to sacrifice one or the other,” he said.
He said the shift also changed how teams work with models and context. Rather than reprogramming systems each time the AI model is tweaked for context, it’s possible to focus on insights instead of worrying about underlying concerns.
“Start with data — monetize your data using AI,” Sitaraman said. “It doesn’t matter how much volume you throw at the initiative, because AI is just truly a linear scale. The more data AI has, the better decisions it makes.”
Still, his warning landed firmly on the difference between automation and autonomy. “One thing we realized is there’s not a lot of difference today between automation and autonomy, and so you have to be careful,” he said.
Sitaraman said professionals have to decide what they actually want: whether they want to automate a process or create an agent. In his framing, creating an agent involves a different cost structure, usage pattern, and governance.
He encouraged teams to identify the right use cases. build the right frameworks. and never underestimate what an agent can do. “You can roll out an agent and say. ‘This is a sales ops agent.’ Often. there’s nothing to stop it from also becoming a sales analyst agent or another type of agent. ” Sitaraman said.
“So, it’s important to ask, ‘Is this what we want it to do?’” he added. “Frameworks are very important, as are skills. You need to think the process through carefully.”
Taken together. the accounts point to a straightforward tension: enterprises want agents to act more autonomously. but governance gaps—especially the ones that surface after incidents—are what push teams toward demotion or decommissioning. The summit leaders didn’t deny that reality. Instead. they described how they’re trying to prevent it by building repeatable evaluation frameworks. backing agents with expert analysts where context is bounded. and focusing on monetizing data while treating autonomy as something that needs governance. not just deployment.
Gartner autonomous AI agents AI governance Snowflake CoCo Whoop Fanatics Synopsys Cortex AI data monetization AI frameworks LLM evaluation