Always-On AI Agents Put Spending Oversight in Focus
Enterprises running Snowflake's coding and task agents must set permissions and track AI spend by team and agent as the tools run unattended.
Updated on Nov. 3, 2026 in Remote Work
Live Poll
Would you trust an AI agent to act on your behalf without asking first?
Snowflake has updated its AI agents Coco, a coding agent for developers, and Cowork, a task agent for general employees, into always-on autonomous systems. The announcement came November 3, 2026 at Snowflake Expedition 2026, and the company also unveiled Cortex AI Gateway and Cortex Training for enterprise agent security, cost control, and custom model building.
Why it matters
Snowflake says AI is shifting from software that waits for prompts to systems that continuously perform tasks, which changes how businesses think about staffing support work and governing AI spend. Operators weighing AI tools now face a question of supervision and cost control, not just access.
Snowflake reports over 9,100 accounts using its Coco coding agent and over 5,800 accounts using its Cowork task agent. Cisco, DocuSign, and Pacific Life Insurance are named as users of both products across their operations, while the source gives no comparison to earlier adoption levels.
The players
Snowflake
Snowflake is a cloud data platform company whose enterprise software and AI tools serve large organizations seeking to manage and act on their data.
Cisco
Cisco is a large networking and enterprise technology company that is named as a customer using both Snowflake agent products across its operations.
DocuSign
DocuSign is a digital agreements and e-signature company named as a customer using both Snowflake agent products across its operations.
Pacific Life Insurance
Pacific Life Insurance is a life insurance and financial services company named as a customer using both Snowflake agent products across its operations.
The details
Once users set broad goals, Coco and Cowork plan and complete tasks in the background without step-by-step prompting. The agents run with the same data access and execution permissions as the user and alert people only for critical decisions or final approvals. The platform automatically routes large-scale background work to lightweight AI models and complex user queries to higher-performance inference models, and the new Cortex AI Gateway acts as a single channel for all enterprise-operated agents.
Timeline
November 3, 2026: Snowflake announces the AI agent updates at Snowflake Expedition 2026.
2026: Snowflake Expedition is held for the first time, expanded from its BUILD developer conference.
Market Landscape
Snowflake's announcement positions its agents against the broader move from prompt-driven AI assistants toward autonomous background task systems. The company is pairing those agents with a central gateway and cost-attribution tooling aimed at enterprise buyers who must control spending across many teams.
Operators evaluating AI agents should weigh supervision requirements, since these agents act with user-level permissions and escalate only critical decisions. Finance and IT teams should plan for per-team and per-agent cost tracking before rolling out always-on tools.
The takeaway
Treat always-on agents as a governance decision as much as a productivity one, and confirm whether your platform can attribute AI spending by team and agent before expanding access. Review how permissions and approval thresholds are configured for any agent that runs in the background.
Further reading
For the broader shift in how teams get work done, see our Remote Work coverage.
Source note: This article includes information reported by 조선일보.
Live Poll
Would you trust an AI agent to act on your behalf without asking first?
