Trust and simplicity are the real AI battlefields

Enterprise tech giants Microsoft and Snowflake debuted tools in San Francisco last week designed to simplify enterprise AI adoption by addressing trust and convenience barriers. Snowflake introduced the Horizon Catalog for data governance and auditable agent identities, while Microsoft unveiled the Agent Control Specification and Project MDASH for security and bug detection. The moves signal that competitive advantage in the AI industry may hinge on providing secure, user-friendly infrastructure rather than developing more powerful models.

AI firms want to remove two of the biggest barriers standing in the way of enterprise adoption: trust and convenience. In San Francisco last week, enterprise tech giants Microsoft and Snowflake both debuted tools aimed at making it easier for companies to adopt and deploy AI, without tedious onboarding processes or concerns about data security. It signals that the winning enterprise AI strategy may not come from the models themselves, but rather the scaffolding that lets enterprises use them. Trust: Snowflake introduced the Horizon Catalog, which serves as a foundational layer for trust and data governance in Snowflake, providing agents with auditable identities, continuous monitoring of their security posture, and a new feature called Horizon Context, which ensures that every agent and person is operating from the same business context. Microsoft, meanwhile, unveiled both Agent Control Specification, an open-source standard that applies controls within the agent loop, and Project MDASH, which uses agents to hunt down exploitable bugs. Simplicity: Microsoft debuted Microsoft Scout, a personal work agent that proactively handles tasks and integrates with tools users commonly use, like Teams and Outlook. Snowflake dropped several updates to Snowflake Cowork, its own work agent, including Cortex Sense to unify data and context, User Memory to learn user behavior to automatically handle tasks, and Skills to create and share ways to automate workflows. These companies are seeking to make it as easy as possible to bring agents into the workforce by handling the most difficult barriers to adoption. Without a foundation of security, cautious enterprises won't feel they can trust the technology, and enterprises that want to move quickly may face a variety of risks if they trust agents unwittingly. "We can't, as an industry, underestimate how much investment is needed to get the trust story right for this technology," Sarah Bird, chief product officer of responsible AI at Microsoft, told The Deep View. However, because the initial guinea pigs for this tech in enterprises were software developers, who are more technically inclined, that may have instilled false confidence in enterprises that deploying AI is easy. If these systems are cumbersome to set up and deploy for tech novices, they may not get as much uptake as they did with coding. "We want the knowledge worker or the business user not to worry about any of the technology, but behind the scenes, to make sure that they have the right MCP connections, they have the right skills, they have the right guardrails," Bala Kasiviswanathan, VP of developer and AI experiences at Snowflake, told The Deep View. "When you make all that happen magically for them, then they will use it more, and they'll trust it more." Our Deeper View These product announcements signal that Microsoft and Snowflake want to meet enterprises where they're at with AI, without sacrificing safety. The fact that major firms recognize this need signals that success in the AI industry may not come from simply training the bigger, better and more powerful model, but rather from selling the tools to leverage them without the headache. In that sense, Microsoft and Snowflake are essentially selling picks and shovels in the AI gold rush. Still, a lot rides on getting it right. This trillion-dollar industry is built entirely on the promise that everyone – enterprises and consumers alike – will eventually adopt this technology. If it's not safe and easy to use, that promise will be delayed or go unfulfilled.