On November 3, 2026, at Snowflake Expedition 2026, Snowflake announced a major update to its two AI agents: CoCo, the coding agent for data teams, and CoWork, the personal work agent for knowledge workers. The headline addition is CoCo and CoWork Bots — always-on AI coworkers that keep working in the background and only bring a person in when a decision or sign-off is needed.
The pitch is a shift from software that waits for a prompt to an autonomous layer that drives continuous execution. Snowflake says more than 14,500 enterprises already run on its AI Data Cloud, with CoCo used by more than 9,100 accounts and CoWork by more than 5,800, measured on average weekly usage across the last four weeks of the quarter ended July 31, 2026. Cisco, Docusign, Klarent, Pacific Life, Shelter Mutual and United Rentals are among the organisations putting both agents to work. The big question is whether always-on agents can be trusted with real work — and Snowflake’s answer is governance first, autonomy second.
What CoCo and CoWork Bots actually do
Until now, AI agents have waited for a specific question or a predefined task before delivering value. CoCo and CoWork Bots change that dynamic entirely. Operating like an autonomous chief of staff, a Bot acts on behalf of the user, inheriting their exact access rights, role context and permissions. Give it a high-level goal — prepare briefings for every customer meeting, or keep data pipelines healthy — and it plans and completes the work in the background, acting directly in the tools teams already use and surfacing only when a critical decision or final approval is needed.
That is the difference between an assistant you query and a coworker you empower. It is also part of a wider shift in how software gets built: AI agents now start 40% of GitHub pull requests, so the work of directing agents, reviewing their output and owning the result is becoming a core skill rather than a novelty.
One important caveat: the Bots themselves are not available yet. Snowflake lists them as coming to private preview soon, so everything about them today is a roadmap promise, not a feature you can switch on.
Automations take recurring work off teams’ plates
While the Bots wait in the wings, the scheduled-work features are here now. With CoCo Automations, generally available in Snowsight and the command-line interface, data teams describe a workflow once and CoCo runs it autonomously at any frequency. Every run keeps an audit trail and task history, so teams can see exactly what CoCo did and suspend or rerun a job whenever they need to — the same discipline that makes a microservices architecture on AWS manageable at scale.
CoWork does the same for business teams by running recurring analysis. A sales leader can ask CoWork to analyse sales performance, identify emerging trends and highlight areas that need attention, then schedule that analysis to run automatically each week, with a fresh report delivered straight to their inbox. A new Async Agent API, also generally available, lets CoWork handle longer, multi-step work that continues running in the background and picks up where it left off if it is interrupted.
AI agents where teams already work: Slack, Excel and mobile
Snowflake is also pushing its agents into the applications where decisions already happen. Snowflake for Slack brings CoCo and CoWork together in one bot experience, so teams get trusted answers in the channels where decisions are made. CoCo for Excel and Sheets brings Snowflake data and plain-language questions into the spreadsheets where finance and operations teams already work, while the data stays protected in Snowflake. And Snowflake Mobile, a unified app for both agents, lets business users ask questions of governed data, get trusted answers and take action with approval, while developers can continue coding sessions, manage builds and steer actions from their phones. All three are in public preview.
The idea is to close the gap between an insight and an action. Instead of losing momentum to tool-switching, a team can move from a question in Slack to a decision in the same thread, with the data never leaving Snowflake’s governed environment.
Governance is the real product
Letting agents work unsupervised raises an obvious problem: an agent that does not know how a company defines revenue, or the rules it runs on, may give a confident wrong answer that someone has to catch. Snowflake’s announcements are mostly an answer to that problem.
Business Ontology, coming to public preview, grounds agents in definitions the business has approved, so two departments cannot get different answers from the same metric. The Cortex AI Gateway governs every model, tool and action agents touch, and its auto model tiers with dynamic model routing — in private preview — match each task to the right model, sending high-volume background jobs to a lightweight model and reserving top-tier reasoning for complex queries, so recurring work does not drive up unnecessary costs.
Skills and Plugin Sharing, generally available, lets users codify CoCo workflows and share them across their organisation with integrated governance, so admins control what counts as official and teams stop rebuilding the same workflow. That matters because agent instructions are shared infrastructure now: GitHub Dependabot now updates AI agent packages through the same pull-request workflow as code, and the same review discipline applies here. Profiles in CoCo and CoWork bring role and team-specific context into the experience, with CoWork Profiles set by admins in a way end users cannot remove. Memory in both agents remembers past context and decisions, with users in control of what is kept.
Snowflake also announced Cortex Training, in public preview, for teams that want models of their own. It tailors open-weight models — in the same direction as Tencent’s Hy4 open-source model — to a company’s domain and proprietary data using techniques like reinforcement learning, serverless, with Snowflake managing the compute. Snowflake’s benchmarks claim up to 61% lower cost than existing serverless training options, a vendor figure to treat as a claim until independently tested. The Snowflake AI Data Package adds more than 60 curated industry datasets with semantic views and a ready-to-use CoWork agent, and Agentic Discovery on Snowflake Marketplace lets users find data, apps and AI through conversation.
What is available now, and what is still coming
It is worth separating the announcement into three buckets, because the most autonomous-sounding pieces are the furthest away. Generally available now: CoCo Automations, the Async Agent API, and Skills and Plugin Sharing. In public preview: Snowflake for Slack, CoCo for Excel and Sheets, Snowflake Mobile, Profiles in CoCo and CoWork, and Cortex Training. In private preview: auto model tiers and dynamic model routing. And coming soon to private preview: the CoCo and CoWork Bots themselves, plus Business Ontology to public preview.
Early customers are already reporting results. Cisco says its customer experience teams use specialised agents built in Snowflake to spot upsell and renewal opportunities, while CoCo helps engineers migrate core corporate data and build pipelines in a fraction of the time it used to take. Docusign says CoCo has cut its data team’s time to deliver value dramatically, and its go-to-market teams can now run deep account research without SQL expertise. United Rentals says work that once took days can now be done in minutes. These are customer quotes from Snowflake’s announcement, not independent measurements, but they show where the company expects the value to land. Channel Insider’s analysis of the launch makes the same caution explicit: partners should treat the Bots as forthcoming, not deployable today.
The bottom line
Snowflake’s bet is that the companies that pull ahead in AI will not be the ones with the cleverest chatbot, but the ones whose agents truly understand their business — its data, its definitions and its rules. CoCo and CoWork Bots are the most ambitious expression of that bet: agents that do not wait to be asked. The Automations, Slack, Excel and mobile integrations are the parts you can use today, and the governance layer is what makes the autonomous parts plausible at all.
Watch three things next: when the Bots actually enter private preview and what they can do there, whether the governance controls hold up once agents act across real enterprise systems, and whether the cost story — dynamic model routing and cheaper custom training — survives contact with production workloads. If those land, the agentic enterprise stops being a conference theme and starts being how work gets done. For the full details, read Snowflake’s official announcement and the Business Wire version of the announcement.
