New: AI API Cost Calculator — compare GPT, Claude, Gemini, Grok & Llama pricing live
Developers · News

Snowflake's Always-On Bots Want to Work While You're Away — Explained Honestly

By ToolsDig Research · November 3, 2026 · How we research

Nothing here is sponsored — no affiliate links on this page. This breaking article is not the daily buyer-intent post. Facts below come from the sources listed at the end, checked 3 November 2026. We have not tested CoCo, CoWork or the Bots hands-on, and nothing here is a hands-on finding.

A calm modern data team workspace at dawn, a laptop and large monitor showing abstract flowing charts and pipeline diagrams in cool blue and teal tones

Most AI assistants still wait for you. You open the chat, you type the question, you get the answer, and the whole thing stops there. Snowflake's pitch at Snowflake Expedition 2026, announced on 3 November 2026, is the opposite shape of work: agents that do not wait to be asked. They run in the background, on a high-level goal you set, and come back to you only when a decision or a sign-off is needed. The agents are called CoCo and CoWork Bots — CoCo for data teams, CoWork for knowledge workers — and Snowflake calls them "always-on."

That phrase deserves a careful reading, because the headline and the availability do not line up in the way a quick skim would suggest. The Bots themselves are not generally available today. What is generally available is a narrower, scheduled piece of the same idea. This article keeps those two things separate throughout, because confusing them is exactly how a launch announcement turns into a buying decision based on something you cannot use yet.

🔍 How we put this together

We verified the story against Snowflake's 3 November 2026 BusinessWire press release (09:06 AM EST) and Snowflake's own blog, with Unite.AI's coverage for the adoption figures, and cross-checked the capabilities, governance model and availability tiers across those reports. We have not run CoCo, CoWork or the Bots ourselves, so nothing here is a hands-on finding. Where a number is the vendor's own — especially the account figures — we label it as exactly that, and where the announcement gives no number, we say so.
🏆 Verdict: Snowflake's always-on Bots are a genuinely interesting idea — agents that work in the background on a goal, inside your existing permissions, and interrupt you only for decisions — and they sit on top of agents that already have thousands of accounts using them. But the headline agent is in private preview soon, not in your hands today. What you can use now is CoCo Automations, and what you can try is the public-preview surfaces. Evaluate those, watch the Bots, and do not buy the keynote version of a product that has not shipped yet.

What it is

CoCo is Snowflake's coding agent for data teams. CoWork is its personal work agent for knowledge workers. The new Bots extend both: instead of answering one prompt at a time, a Bot works proactively in the background on a high-level goal — Snowflake's examples are preparing briefings for every customer meeting, and keeping data pipelines healthy. The Bot plans the work, completes it, and surfaces only when a decision or sign-off is needed.

The governance claim is the part Snowflake leans on hardest, and it is also the part that matters most if you work with governed data. The Bots inherit the user's exact access rights, role context and permissions. In other words, the agent does not get a master key to the warehouse: it sees what you see, in the role you hold. That is the "grounded in governed data" half of the announcement's framing — the agent's work is supposed to stay inside the access rules your organisation already set, rather than routing around them.

Alongside the Bots, Snowflake announced a cluster of related pieces: CoCo mode in Snowsight, Cloud Agents, CoWork Profiles, dashboards in CoWork, Memory and Artifacts, and Skills and Plugins Sharing, which is generally available. In public preview are Snowflake for Slack, CoCo for Excel and Sheets, and Snowflake Mobile, a unified CoCo and CoWork app. The announcement's shape is therefore a platform widening in several directions at once — chat, spreadsheets, mobile, dashboards — with the Bots as the headline idea sitting on top of it.

What it actually does today — and what is only preview

This is the distinction the whole article turns on, so here it is plainly:

That ordering matters for planning. If your team wants scheduled recurring data work inside Snowflake today, CoCo Automations is the real thing to evaluate. If you want a background agent that prepares every meeting briefing unprompted and only interrupts you for sign-off, that is the private-preview promise — worth watching, not worth building a workflow around yet.

The line Snowflake drew — and why it matters

The most reassuring sentence in the announcement is not about capability at all. It is about permission: the Bots inherit the user's exact access rights, role context and permissions. For data teams, that is the difference between an agent that is useful and an agent that is a governance incident waiting to happen. An agent that plans and completes work in the background, across governed data, is only acceptable if it cannot see more than the person who set it running. Snowflake is saying it cannot — and that claim, if it holds in practice, is what makes the background-work idea deployable inside a real enterprise rather than just impressive in a keynote.

The second line is the availability line, and Snowflake has at least drawn it honestly in its own blog: "private preview soon" for the Bots, public preview for the new surfaces, general availability for Automations and for Skills and Plugins Sharing. The risk is not in Snowflake's wording; it is in how quickly "announced" becomes "available" in second-hand coverage. Keep the three tiers separate and the announcement reads as a credible roadmap with one solid shipped piece. Blur them and it reads as a product you can buy today that does not exist yet in that form.

The vendor claims — labelled as the vendor's

Snowflake's adoption figures, as reported by Unite.AI, are that CoCo is used by more than 9,100 accounts and CoWork by more than 5,800 accounts, averaged over the last four weeks of the quarter ended 31 July 2026. Named users include Cisco, Docusign and United Rentals. Those are Snowflake's own claims, reported by others, not numbers we have tested or verified independently. Treat them as evidence that CoCo and CoWork already have real usage behind them — the Bots are an extension of agents people already use, not a launch from zero — not as independently measured results.

On pricing, the announcement gives no numbers, so this article gives none either. Nothing here should be read as implying a price, a tier, or a cost model for the Bots, CoCo or CoWork. If cost matters to your decision — and for a platform that touches your data warehouse, it should — that is a question to put to Snowflake directly, not something to guess from a launch announcement.

Who it matters for

The limits — stated plainly

The bottom line is a measured one. Snowflake's always-on Bots are a genuinely interesting idea — agents that work in the background on a goal, inside your existing permissions, and interrupt you only for decisions — and they sit on top of agents that already have thousands of accounts using them. But the headline agent is in private preview soon, not in your hands today. What you can use now is CoCo Automations, and what you can try is the public-preview surfaces. Evaluate those, watch the Bots, and do not buy the keynote version of a product that has not shipped yet.

Sources: Snowflake press release via BusinessWire, 3 November 2026, 09:06 AM EST: "Snowflake CoCo and CoWork Power the Agentic Enterprise, Grounded in Governed Data and Built for Autonomous Action" (BusinessWire); Snowflake blog: "AI agents, governed data, business context" (Snowflake); Unite.AI: "Snowflake Unveils Always-On CoCo and CoWork Bots for Background Work" (Unite.AI). Adoption figures are Snowflake's claims as reported, not independently tested results. No hands-on testing was performed for this article.