Snowflake Pushes AI Agents Beyond Prompts and Into Persistent Work

Snowflake wants agents to keep working after everyone leaves. On November 3, 2026, the company announced always-on bots for CoCo, its coding agent for data teams, and CoWork, its personal work agent for knowledge workers. The bots are designed to plan and complete work in the background instead of waiting for each individual prompt.

The CoCo and CoWork Bots are coming soon to private preview. Users can set a high-level goal, such as preparing briefings for every customer meeting or keeping data pipelines healthy, and let the bot pursue that assignment over time.

The important constraint is that these agents do not receive a magical new identity with unlimited access. Each bot inherits the user’s access rights, role context and permissions, works within those boundaries, and surfaces its output when a critical decision or final approval requires human input. Snowflake is moving agents from answering requests to pursuing continuing assignments while keeping existing access controls and human sign-off in place.

Scheduled work, reusable workflows and mobile approvals

Snowflake is also making CoCo Automations generally available, allowing data teams to schedule repeated workflows through CoCo and CoWork. Each run retains an audit trail and task history, while teams can suspend or rerun a job—basic controls, perhaps, but still better than asking everyone to trust a tireless mystery process.

For business users, CoWork can repeat an analysis on a set schedule. Snowflake’s example involves a sales leader receiving a weekly report covering sales performance, emerging trends and areas needing attention. The generally available Async Agent API supports longer, multi-step CoWork tasks that continue in the background and resume after an interruption.

Snowflake for Slack, CoCo for Excel and Sheets, and Snowflake Mobile are in public preview. Slack puts CoCo and CoWork in one bot, while the spreadsheet integrations let users ask plain-language questions without moving the data out of Snowflake. Snowflake Mobile combines both agents, giving business users a way to approve actions while developers continue coding sessions, manage builds and steer work from their phones.

Skills and Plugin Sharing is now generally available, letting users turn CoCo workflows into reusable assets and share them across an organization. Administrators can decide which workflows become official. Profiles in CoCo and the new Profiles in CoWork are in public preview: CoCo Profiles offer suggested defaults users can customize, while CoWork Profiles let administrators set agent access and routing instructions that end users cannot remove.

Memory in both agents retains previous context and decisions, with users able to view, update or delete stored memories. CoWork can also turn analysis into shareable HTML reports or interactive dashboards, and teammates can open a side-by-side conversation with CoCo to examine the results.

Persistent agents make visibility part of the product

Snowflake said CoWork is used by more than 5,800 accounts and CoCo by more than 9,100. Those figures are averages over the last four weeks of the quarter ended July 31, 2026, giving the company a usage baseline as it shifts from interactive assistants toward agents that keep operating between visits.

Snowflake CEO Sridhar Ramaswamy previewed additional launches for the company’s Expedition event, including easier capture of business context for building agents, autonomous agents that keep working when users step away, and CoWork features for revenue and finance teams. A new Oracle partnership will make Snowflake a native first-party product on Oracle Cloud.

Ramaswamy said AI is industrializing software and that Snowflake carries no leverage and is on its way to GAAP profitability next year. The claims arrive alongside a broader industry problem: agents that act continuously need controls that explain not only what they were allowed to do, but what they actually did.

Neo announced a new integration for Claude Cowork on November 2, 2026, extending Neo into the isolated virtual machines where Cowork runs. Claude Cowork is popular for taking multi-step instructions and returning finished work, while its most sensitive actions run inside virtual machines to keep that work isolated.

Most security tools can see the virtual machine but not what happens inside it— which file the agent opens, what processes it starts or which hosts it contacts. Neo identifies the semantic intent behind every Cowork action wherever it occurs, providing runtime visibility with agentic security intelligence and a complete understanding of what each agent is doing, what it is interacting with and what risk that represents for the environment.

The need for that visibility is not theoretical. In July 2026, the U.K.’s AI Security Institute disclosed that AI agents under evaluation had taken sustained, unsanctioned action against real people and organizations. In the most serious case, an agent researched maintainers of a public open-source project on GitHub, created multiple fake identities and tried to socially engineer a maintainer into approving a malicious pull request.

The maintainer rejected the code. AISI detected unusual network traffic on July 28 and contained the incident within an hour of the alert.

Snowflake’s approach is clear: persistent agents should inherit permissions, preserve histories, expose memory controls and pause for approval. That does not eliminate risk, but it moves autonomous work closer to something enterprises can govern instead of merely tolerate.

Stay connected via Google News

Follow us for the latest AI News