Snowflake CoCo and CoWork Explained: Always-On AI Agents That Keep Working After You Log Off

Snowflake CoCo CoWork is the biggest change to enterprise AI agents in 2026 so far. At Snowflake Expedition 2026 on 3 November 2026, Snowflake announced that its two AI agents — CoCo, the coding agent for data teams, and CoWork, the personal work agent for knowledge workers — will stop waiting for a prompt and start working on their own. The new Snowflake CoCo CoWork Bots are designed to keep working in the background, finish recurring jobs, and only come back to a person when a decision or a sign-off is needed.

That is a real shift, and it is also a launch that needs careful reading. Some of what Snowflake announced is generally available today. Some of it is in public preview. And the headline feature — the always-on Bots themselves — is still coming soon to private preview. This guide explains exactly what Snowflake CoCo CoWork is, what each piece does, which parts you can use right now, what it costs, and who should actually care.

Snowflake CoCo CoWork enterprise data center powering always-on AI agents

Table of Contents

What Snowflake CoCo CoWork Actually Is

Snowflake CoCo CoWork is not one product. It is a pair of AI agents that live inside the Snowflake AI Data Cloud, and the two agents serve two different audiences. CoCo is the coding agent for data teams: it writes and fixes data pipelines, converts legacy code, builds dashboards from a description, and now runs scheduled data work on its own. CoWork is the personal work agent for knowledge workers: it answers business questions in plain language, runs recurring analysis, and turns answers into reports and dashboards a whole team can use.

The idea behind both agents is simple, and it is the same idea Snowflake has been building toward all year. Most AI agents answer a question and then stop. Snowflake wants the Snowflake CoCo CoWork agents to keep going — to take a goal, plan the work, run it on a schedule, and finish it. At Expedition 2026, Snowflake’s EVP of Product, Christian Kleinerman, put the bet in one line: the difference between agents that guess and agents that know comes down to data. Snowflake already holds the data, the business context, and the governance controls that enterprises run on, so its agents start with an advantage most standalone chatbots never get.

The numbers Snowflake shared show the scale of that bet. More than 14,500 global enterprises use Snowflake, and the company says CoCo is now used by more than 9,100 accounts while CoWork is used by more than 5,800 accounts. Those figures are averages over the last four weeks of the quarter ended 31 July 2026, counted from capacity and on-demand accounts using each feature weekly. They are Snowflake’s own numbers, not independently verified figures, but they show that both agents already have a large installed base before this expansion.

What changed at Expedition 2026 is the mode of work. Until now, the Snowflake CoCo CoWork agents were interactive: you asked, they answered. The new release adds three layers on top of that foundation. First, scheduled and recurring work through Automations. Second, always-on Bots that take a high-level goal and work on it in the background. Third, a governance foundation — Business Ontology and Cortex AI Gateway — that is meant to stop agents from confidently answering with the wrong definition of revenue, or the wrong access to a sensitive table. Each layer matters, and each layer has a different availability status, which is why this guide keeps them separate.

CoCo and CoWork Bots: Always-On AI Coworkers

The headline of the Snowflake CoCo CoWork announcement is the new Snowflake CoCo CoWork Bots. Snowflake describes them as always-on AI coworkers that operate like an autonomous chief of staff. You give a Bot a high-level goal — prepare a briefing for every customer meeting, or keep our data pipelines healthy — and the Bot plans the work, completes it in the background, and surfaces only when a critical decision or a final approval is needed.

That is a genuinely different promise from a chatbot. A chatbot waits. A Bot, in Snowflake’s framing, starts. It works while you are in another meeting, and it keeps working after you log off. When it finishes, it brings you the result instead of making you go and ask for it.

The most important design choice is about permissions. A Bot acts on behalf of the user, and it inherits that user’s exact access rights, role context, and permissions. It does not get a new, powerful identity of its own. It cannot see data the user cannot see, and it works strictly within the user’s governed boundaries in Snowflake. That constraint is what makes the always-on idea deployable inside a large company at all: the agent is powerful, but it is powerful only inside the box the business already drew around that employee.

Here is the honest caveat, and it is the one that matters most for anyone planning a rollout. The CoCo and CoWork Bots are not generally available. Snowflake marks them as private preview soon — limited, forthcoming access, not something you can switch on today. Several independent write-ups of the announcement made the same point in the days after Expedition: the Bots are the furthest step in this release, and they are not a general release. If a vendor or a partner tells you the Bots are live, ask them to show you the private preview enrolment, because the public announcement says otherwise.

That does not make the announcement empty. The Bots sit on top of features that are shipping, and understanding those features is the practical way to prepare for the Bots when they arrive.

CoCo Automations: Scheduled Work That Is Live Today

CoCo Automations is the part of the Snowflake CoCo CoWork release you can actually use now. It is generally available in Snowsight and the command-line interface, with public preview in CoCo Desktop. The concept is straightforward: a data team describes a workflow once, and CoCo runs it autonomously at whatever frequency the team sets.

Typical examples from Snowflake’s announcement include pipeline health checks and weekly reports. A data engineer can ask CoCo to check a set of pipelines every morning, flag anything that broke overnight, and post a summary. A team can schedule a recurring data-quality check, a regular refresh of a reporting table, or a weekly roll-up that used to need a manual trigger. Every run keeps an audit trail and task history, so teams can see exactly what CoCo did, and they can suspend or rerun a job whenever they need to.

The Snowflake CoCo CoWork release does the same thing for business teams through CoWork. A sales leader, for example, can ask CoWork to analyse sales performance, identify emerging trends, and highlight areas that need attention — and then schedule that analysis to run automatically each week, with a fresh report delivered straight to their inbox. The work is recurring, but the request is made once.

Two related pieces support the scheduled-work story. The 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. And CoCo Cloud Agents, announced in the companion autonomous data engineering release, run CoCo in an isolated environment that Snowflake manages instead of on an engineer’s laptop — also generally available. Together, Automations, the Async Agent API, and Cloud Agents are the parts of the Snowflake CoCo CoWork platform that turn an interactive assistant into a worker that keeps going when nobody is watching.

There is one more generally available piece in this cluster: Skills and Plugin Sharing. It lets users codify a CoCo workflow into a reusable skill and share it across the organisation, with integrated governance. Administrators decide which workflows are designated official, which standardises processes and saves tokens, because teams stop rebuilding the same workflow from scratch. When teams build AI in silos, the same workflows get rebuilt and the same mistakes repeat; shared skills are Snowflake’s answer to that waste.

Snowflake CoCo CoWork coding agent helping a data engineer write code

CoWork for Business Teams: Profiles, Dashboards and Automations

CoWork, the business-facing half of the Snowflake CoCo CoWork platform, got its own set of upgrades at Expedition 2026. Two new capabilities — Profiles and Dashboards — sit alongside Automations and take a business team from a shared report to finished, recurring work.

Profiles give each team the right configuration for its agent. In CoWork, a Profile lets administrators set agent access and routing instructions that end users cannot remove, so a finance team and a sales team can each get an agent tuned to their own data, tools, and rules. Profiles never widen access: CoWork runs with each user’s own privileges, and a Profile will not route a question to an agent that user is not allowed to use. Profiles in CoCo and the new Profiles in CoWork are in public preview.

Snowflake Dashboards, also in public preview, take an AI-first approach to business intelligence. An analyst builds a Dashboard in the Snowsight interface by describing it to Snowflake CoCo, then shares it with the right roles so business users can access it. Business users open the Dashboard in CoWork, or optionally in Snowsight. One person’s analysis becomes something the whole team can discover and explore, whether it is a polished HTML report or an interactive dashboard.

Underneath Profiles and Dashboards, CoWork keeps the capabilities that make it a work agent rather than a question-answering bot. Memory remembers past context and decisions so nobody has to explain the same thing twice, and users control what is kept — they can view, update, or delete stored memories at any time. Analytical search, still in preview, lets CoWork answer questions that span thousands of documents, such as which customer issues came up most last month: it narrows the set with Cortex Search, then uses AI functions and SQL to count and compare, and shows its plan before it runs. Document generation, code execution, and connectors turn answers into finished work a team can share.

Snowflake is also welcoming the Mora team to help bolster its AI-for-BI efforts, incorporating their approach into the platform. Teammates can open a side-by-side conversation with CoCo to dig deeper into what they are seeing, or build internal enterprise apps with Snowflake Apps for advanced exploration or multi-step workflows.

Where the Agents Show Up: Slack, Excel, Sheets and Mobile

An agent that only lives inside a data platform will only be used by people who already live inside that platform. Snowflake knows that, and the Expedition 2026 release pushes the Snowflake CoCo CoWork agents into the tools teams already use every day. Three of those integrations are in public preview, and the distinction matters for planning.

Snowflake for Slack brings CoCo and CoWork together in one bot experience inside Slack, so teams get trusted answers in the channels where decisions are actually made. Instead of leaving a conversation to check a number, a team can ask the question where the discussion is happening and get an answer grounded in governed Snowflake data.

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. That last clause is the point: the spreadsheet is the interface, but Snowflake remains the system of record, so the numbers stay governed even when the work happens in Excel or Google Sheets.

Snowflake Mobile is a unified app that brings CoCo and CoWork together on a phone. Business users can 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. Users who rely on both agents can switch between the two experiences seamlessly, all within Snowflake’s existing security and governance controls.

Snowflake CoCo CoWork mobile app letting business users ask governed data questions on the go

The mobile angle connects directly to the always-on Bots story. A Bot that works in the background is only useful if a person can check in on it, approve its work, and steer it from wherever they are. Snowflake Mobile is the control surface for that idea, and its public preview status means teams can start testing the workflow now, before the Bots themselves arrive.

The Governance Layer: Business Ontology and Cortex AI Gateway

The most consequential part of the Snowflake CoCo CoWork release may not be the agents at all. It is the governance layer underneath them. Snowflake’s own framing is blunt: 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. In a large enterprise, that confident wrong answer can travel a long way before anyone notices.

Snowflake cites a report from MIT Technology Review that it sponsored, which found that 55% of executives named lack of business context as the main reason for low-quality agent insights. Whether or not that figure generalises, the problem it describes is real: every department defines the same metric slightly differently, and an agent that guesses which definition to use will eventually guess wrong.

Business Ontology: One Definition of Every Metric

Business Ontology, a new capability inside Snowflake Horizon Catalog, is Snowflake’s answer. It gives AI agents access to approved definitions of business terms and links them to the underlying enterprise data, so agents and employees use consistent definitions for metrics such as net revenue. Instead of each team maintaining its own spreadsheet of definitions, the ontology becomes the single source of truth the agents consult before they answer.

Business Ontology builds on Snowflake’s existing semantic modelling capabilities, and analysts who covered the launch called it perhaps the most valuable feature of the whole Expedition release. It is coming soon to public preview — not generally available yet, but close enough that teams can start preparing their definitions now.

Cortex AI Gateway: A Checkpoint Every Agent Must Pass

Cortex AI Gateway is the control layer. It governs every model, tool, and action an agent touches, applying access policies, authentication, and permissions across AI providers and frameworks. It includes fine-grained authorisation, protection against prompt injection and jailbreaks, cost governance, dynamic model routing, and observability of agent activity. In practice, it is the checkpoint every AI agent in a company must pass through before it can act.

One feature inside the Gateway deserves attention on its own: auto model tiers with built-in dynamic model routing for CoCo. It lets teams dynamically match every task to the right model, assigning high-volume background jobs to a lightweight model while reserving top-tier reasoning for complex queries. Each task runs at the optimal balance of performance and price, which matters enormously once agents start running recurring work on a schedule — because scheduled work multiplies token usage, and unmetered scheduled work multiplies cost. Dynamic model routing is in private preview.

The governance story is also why the Snowflake CoCo CoWork announcement pairs naturally with the wider debate about AI agents in the enterprise. Other companies are blocking agents outright — our guide to why websites block AI agents explains how Yelp, Delta, JPMorgan and Schwab have shut agents out — while Snowflake is betting the opposite way: let agents in, but only through a governed checkpoint, with the user’s own permissions, and with an audit trail for everything they do. It is the same tension from the other side, and the market is still deciding which approach wins.

Snowflake CoCo CoWork governance meeting reviewing AI agent access and controls

Cortex Training: Cheaper Custom Models for Your Own Data

Alongside the Snowflake CoCo CoWork agent news, Snowflake announced Cortex Training, in public preview. It lets organisations build AI models that are uniquely their own, tailored to their domain and proprietary data with techniques like reinforcement learning, in a serverless manner while Snowflake manages the computing power and reduces costs with GPU multiplexing. Every step, from preparing training data to deploying the finished model, happens inside Snowflake’s governed environment.

The pitch is aimed at a real cost problem. Agents built on frontier models get expensive fast once they scale, because every step and tool call burns tokens. Growing pressure around AI sovereignty is also pushing companies toward open-weight models for more control and lower costs. But open-weight models are general-purpose out of the box, and customising them has usually meant waiting months for scarce GPU capacity and managing complex infrastructure. Cortex Training removes that burden, according to Snowflake.

The headline number is striking: Snowflake says benchmarks show up to 61% lower cost than existing serverless training options. Read the footnote before repeating that number, though. Snowflake’s pricing comparison is based on pricing published on 22 October 2026 on the Fireworks and Tinker webpages, running LoRA SFT training workloads on Cortex Training and measuring throughput per H200 GPU, then calculating price per million tokens from throughput. It is a vendor benchmark with a stated methodology, not an independent test, and your mileage will depend on your workload. Treat 61% as Snowflake’s claim, not as a guarantee.

Cortex Training is also built on an open source foundation, so the broader AI community can see how it works and help improve it. One early user, Poetiq, said it is using Cortex Training to post-train the first large language model in the world that specialises in optimisation, handling everything from LoRA to SFT to full-parameter post-training with a custom GRPO implementation.

Two more data-facing pieces round out the release. The new Snowflake AI Data Package, in public preview, offers more than 60 curated industry datasets, complete with semantic views and a ready-to-use CoWork agent. Agentic Discovery on Snowflake Marketplace, also in public preview and powered by CoCo, lets users find data, apps, and AI on Snowflake Marketplace through conversation, then install, trial, or request listings right from the chat. Providers can now generate a Cortex Agent and a semantic view for their data and attach them to a listing, so recipients can ask questions in plain language instead of writing queries.

What Is Generally Available and What Is Still in Preview

The single most useful thing in this guide is the availability table below. The Snowflake CoCo CoWork announcement mixes generally available features, public previews, private previews, and one feature that is only coming soon to private preview. Treating them all as “launched” is the easiest mistake a reader — or a buyer — can make.

Feature Status at Expedition 2026 What it means
CoCo Automations Generally available Schedule recurring data workflows in Snowsight and the CLI; public preview in CoCo Desktop.
Async Agent API Generally available Long, multi-step CoWork tasks run in the background and resume after interruption.
Skills and Plugin Sharing Generally available Turn CoCo workflows into reusable, governed skills shared across the company.
CoCo Cloud Agents Generally available CoCo runs in an isolated, Snowflake-managed environment instead of a laptop.
CoCo Restricted Session Scope Generally available Tighter session boundaries for autonomous data engineering work.
Profiles in CoCo and CoWork Public preview Role and team-specific agent configuration; admins control CoWork routing.
Snowflake Dashboards Public preview AI-built dashboards described to CoCo, shared by role, opened in CoWork.
Snowflake for Slack Public preview CoCo and CoWork in one Slack bot experience.
CoCo for Excel and Sheets Public preview Plain-language Snowflake questions inside spreadsheets.
Snowflake Mobile Public preview Unified CoCo and CoWork app for phones.
Cortex Training Public preview Serverless post-training of open-weight models on your own data.
AI Data Package Public preview 60+ curated industry datasets with semantic views and a CoWork agent.
Agentic Discovery on Marketplace Public preview Find and try data, apps, and AI through conversation.
Business Ontology Public preview soon Approved business-term definitions for agents; not open yet.
Auto model tiers / dynamic model routing Private preview Match each task to the right model automatically, inside Cortex AI Gateway.
CoCo and CoWork Bots Private preview soon The always-on coworkers; limited, forthcoming access only.

The pattern is clear. The scheduled, auditable work is what Snowflake is shipping now. The always-on, autonomous work is what Snowflake is promising next. That sequencing is deliberate: scheduled work can be reviewed, suspended, and rerun, which makes it safe to ship. A Bot that plans and completes whole assignments on its own needs the governance layer — Business Ontology, Cortex AI Gateway, restricted session scope — to be solid first, and some of that layer is still in preview itself.

Who Is Already Using Snowflake CoCo CoWork

Snowflake named six organisations putting CoCo and CoWork to work: Cisco, Docusign, Klarent, Pacific Life Insurance Company, Shelter Mutual Insurance Company, and United Rentals. Their quotes in the announcement give a useful sense of what the agents are actually doing in production, rather than what they might do one day.

At Cisco, customer experience teams use specialised agents built in Snowflake to identify opportunities for smarter upsells and renewals, while CoCo helps engineers accelerate the migration of core corporate data — converting legacy code and building pipelines in a fraction of the time it used to take. Srini Namineni, Cisco’s SVP and Chief Automation Officer, framed it as managing complexity at global scale: vast amounts of data across multiple business units and architectures.

At Docusign, CoCo and CoWork cut the data team’s time to deliver value dramatically, according to Chief Data Officer Bala Sahejpal, and power the company’s conversational data agents. Go-to-market teams can now run deep account research and get insights into an account without needing SQL expertise.

At United Rentals, the world’s largest equipment rental company, CoCo and CoWork are helping transform how teams work. Tony Leopold, SVP and Chief Technology and Strategy Officer, said that what once took days can now be done in minutes, with employees using natural language to understand their data, discover actionable insights, and make more informed decisions.

These are early-adopter stories from large enterprises, and they should be read that way. None of them describes the always-on Bots, because the Bots are not in private preview yet. What they describe is the interactive and scheduled-work layer that is already live — and that layer is clearly delivering real time savings for data teams and business teams alike.

Snowflake CoCo CoWork business team collaborating with AI agent insights

What Snowflake CoCo CoWork Costs

Snowflake did not announce a separate price for CoCo, CoWork, or the Bots at Expedition 2026, and there is no per-seat price to quote. That is not an oversight; it is how Snowflake sells. Snowflake’s platform is consumption-based: customers pay for the compute and storage they use, and AI features such as Cortex consume credits on top of that. The cost of running Snowflake CoCo CoWork therefore depends on how much work the agents do — how many queries they run, how much data they touch, and which models they use.

That is exactly why the cost-governance pieces in this release matter so much. Dynamic model routing inside Cortex AI Gateway exists to keep recurring agent work from driving up unnecessary costs: high-volume background jobs go to a lightweight model, and only complex queries get top-tier reasoning. Skills and Plugin Sharing saves tokens by reusing proven workflows instead of rebuilding them. Cortex Training’s claimed 61% cost saving is aimed at the same problem from the model side.

For a US business evaluating the platform, the honest budgeting advice is this: ask your Snowflake account team for a Cortex consumption estimate based on your expected agent usage — number of users, scheduled automations, and the models you plan to route to — before committing to a rollout. The agents themselves are not a line item with a sticker price; the bill is the compute they consume, and scheduled, always-on work consumes compute even when nobody is at a keyboard. That is a feature, not a bug, but it has to be budgeted like one.

There is also a practical cost to the preview status of the Bots and the governance layer. Teams that adopt the generally available Automations today will be in the best position to switch on Bots later, because the workflows, skills, and audit trails they build now are the same ones the Bots will use. Teams that wait for the Bots to go generally available will start from zero on that preparation.

How Snowflake CoCo CoWork Compares to Other AI Agents

Snowflake is not the only company pushing agents from answering questions into doing work, and the Snowflake CoCo CoWork launch lands in a crowded field. The comparison that matters most for a buyer is not feature-by-feature; it is where the agent lives and what it is allowed to touch.

Versus general-purpose coding and work agents: tools such as GitHub Copilot, Claude Code, and the new wave of multi-agent coding orchestrators are excellent at writing code and running developer workflows. Snowflake CoCo CoWork is different because it is data-native: it lives where the governed data already lives, it understands the company’s semantic models, and it inherits the user’s Snowflake permissions. A general coding agent can write a SQL query; CoCo can run it against governed data, schedule it, and hand the result to a business user in Slack. That is a narrower scope with a deeper integration.

Versus enterprise agent platforms: Zoho’s agentic analytics, which we covered in our Zoho Analytics Agentic Analytics guide, brings agents to business intelligence inside Zoho’s suite. Snowflake’s answer is broader: CoCo and CoWork are the agents, but Business Ontology and Cortex AI Gateway are the control plane that any agent — including third-party ones — must pass through. Snowflake is positioning itself as the governed foundation for the agentic enterprise, not just as another agent vendor.

Versus the agent-shopping wave: the same week Snowflake announced its Bots, the consumer side of agents was moving in the opposite direction in some places and the same direction in others. Walmart’s Agent Gateway lets AI assistants shop and buy on a customer’s behalf, with spending limits and approval controls — a consumer version of the same “agent acts, human approves” pattern. Meanwhile, other companies are blocking agents outright. Snowflake’s bet is that the enterprise answer is neither a ban nor an open door, but a governed checkpoint.

Versus Claude Cowork: Anthropic’s Cowork is a closer cousin — a personal work agent that runs in an isolated virtual machine. In fact, on 2 November 2026, Neo announced an integration extending into the isolated virtual machines where Cowork runs. The difference is scope: Claude Cowork is a general-purpose agent for a single user’s computer, while Snowflake CoCo CoWork is an enterprise agent pair tied to governed company data, shared skills, and organisation-wide audit trails. A freelancer might pick Cowork; a 10,000-person company with a Snowflake estate will look at CoCo and CoWork first.

Should Your Team Use It — or Wait?

The answer depends on where your team sits today, and the Snowflake CoCo CoWork availability table above is the decision tool.

Use CoCo Automations now if your data team already runs on Snowflake and repeats the same checks, refreshes, or reports every week. Automations are generally available, audited, and reversible. Start with one scheduled workflow — a daily pipeline health check is the classic first automation — and measure the hours it saves before adding more.

Try the public previews if your business teams live in Slack, Excel, or on their phones. Snowflake for Slack, CoCo for Excel and Sheets, and Snowflake Mobile are all in public preview, and they are the fastest way to find out whether business users will actually talk to their data in plain language. Roll them out to one team first, and watch whether the questions people ask get better over time.

Prepare for the Bots, but do not plan a launch around them yet. The always-on CoCo and CoWork Bots are coming soon to private preview, with no public date. The preparation that pays off is unglamorous: clean up your business definitions so Business Ontology has something true to work with, codify your best CoCo workflows into shared skills, and set up the audit habits that let you review what an agent did. When the Bots arrive, the teams that did that work will be ready; the teams that waited will be starting from scratch.

Wait, or at least go slowly, if your Snowflake governance is not yet in order. An agent that inherits a user’s permissions is only as safe as those permissions are. If your roles are over-broad, or your metric definitions differ between departments, an always-on agent will amplify those problems at machine speed. Fix the governance first; the agents will still be there when you are ready.

One more caution applies to every enterprise agent announcement, including this one. The usage figures, the cost-saving benchmarks, and the customer quotes all come from Snowflake or its customers. They are useful signals, not independent verification. The right way to evaluate the Snowflake CoCo CoWork platform is the same as for any enterprise platform: run a bounded pilot on your own data, measure the hours saved and the credits consumed, and let your own numbers decide.

Frequently Asked Questions

What is Snowflake CoCo CoWork?

Snowflake CoCo CoWork is the pair of AI agents inside the Snowflake AI Data Cloud: CoCo, the coding agent for data teams, and CoWork, the personal work agent for knowledge workers. At Expedition 2026 on 3 November 2026, Snowflake added scheduled Automations, always-on Bots, and a governance layer built on Business Ontology and Cortex AI Gateway.

Are the Snowflake CoCo CoWork Bots available now?

No. The always-on CoCo and CoWork Bots are coming soon to private preview, which means limited, forthcoming access — not something you can switch on today. CoCo Automations, the Async Agent API, and Skills and Plugin Sharing are generally available now, while Slack, Excel, Sheets, and Mobile integrations are in public preview.

What is the difference between CoCo and CoWork?

CoCo is the coding agent for data teams: it writes and fixes pipelines, converts legacy code, and builds dashboards. CoWork is the personal work agent for knowledge workers: it answers business questions in plain language, runs recurring analysis, and turns answers into reports and dashboards. Both inherit the user’s exact Snowflake permissions.

How much does Snowflake CoCo CoWork cost?

There is no separate per-seat price. Snowflake is consumption-based, so CoCo and CoWork consume Cortex credits based on the compute and models they use. Ask your Snowflake account team for a Cortex consumption estimate based on your expected usage, and use dynamic model routing to keep scheduled work on cheaper models where possible.

What is Snowflake Business Ontology?

Business Ontology is a capability inside Snowflake Horizon Catalog that gives AI agents access to approved definitions of business terms — such as net revenue — linked to the underlying data. It is meant to stop agents from using the wrong definition of a metric. It is coming soon to public preview.

Conclusion

Snowflake CoCo CoWork is the clearest sign yet that enterprise AI agents are moving from answering questions to doing work — on a schedule, in the background, and inside the tools teams already use. The generally available pieces are genuinely useful today: CoCo Automations, the Async Agent API, and Skills and Plugin Sharing turn recurring data work into something that runs itself, with an audit trail for every run. The public previews — Slack, Excel, Sheets, Mobile, Profiles, Dashboards, and Cortex Training — show where the platform is heading next.

But the honest summary is the availability table. The headline feature, the always-on Bots, is still coming soon to private preview, and some of the governance layer it depends on is still in preview too. Snowflake’s bet is that governed, data-native agents will win over standalone chatbots, and the early evidence from Cisco, Docusign, and United Rentals supports that bet. The smart move for most teams is to start with what is live — one scheduled automation, one team on the Slack or mobile preview — and build the definitions, skills, and audit habits that the Bots will need when they arrive. That way, when the always-on coworkers finally show up, they will be working for a team that is ready for them.

Sources: Snowflake’s official press release, Snowflake CoCo and CoWork Power the Agentic Enterprise, 3 November 2026; Snowflake’s CoWork Profiles, Dashboards and Automations blog post; TechTarget and Channel Insider coverage of Snowflake Expedition 2026.

Leave a Comment

Your email address will not be published. Required fields are marked *