Snowflake adds always-on AI bots for CoCo & CoWork

Snowflake has introduced new features for its CoCo and CoWork AI agents, adding always-on bots that can carry out recurring tasks in the background.

The changes are intended to narrow the gap between user requests and the business context agents need to act on them.

CoCo is Snowflake's coding agent for data teams, while CoWork is aimed at knowledge workers. The latest release adds bots that can operate continuously, use a customer's existing permissions and role settings, and seek approval only when a decision or sign-off is required.

Snowflake is also adding automation features so teams can schedule repeated workflows. For data teams, that includes automated runs through CoCo with an audit trail and task history. Business teams can use CoWork to schedule recurring analysis and reports.

The tools will work across software employees already use, including Slack, Excel, Sheets and mobile devices. Snowflake is also introducing an asynchronous application programming interface for CoWork so longer, multi-step tasks can continue in the background and resume if interrupted.

Usage figures released by Snowflake suggest the products already have a sizeable installed base. CoWork is used by more than 5,800 accounts and CoCo by more than 9,100. Snowflake said it serves more than 14,500 enterprise customers globally.

Christian Kleinerman, Executive Vice President of Product at Snowflake, said the wider issue for businesses is whether AI agents have enough access to reliable internal context to avoid confident but inaccurate answers.

"AI agents are reshaping how companies operate, and the difference between agents that guess and agents that know comes down to data," said Christian Kleinerman, Executive Vice President of Product at Snowflake.

"Snowflake already holds the data, context and governance controls that enterprises run on. With Snowflake CoCo and CoWork, we're turning that foundation into agents that understand the business and get better every day. This is the agentic enterprise, and it's built on Snowflake."

Customer use

Snowflake pointed to examples from customers including Cisco, Docusign and United Rentals to show how the software is being applied in data migration, account research and operational analysis.

At Cisco, internal teams use agents built in Snowflake for customer upsell and renewal work, while engineers use CoCo to convert legacy code and build data pipelines.

"Operating at Cisco's global scale means managing vast amounts of data across multiple business units and architectures," said Srini Namineni, Senior Vice President and Chief Automation Officer at Cisco.

"Our Customer Experience teams use specialised agents built in Snowflake to identify opportunities for smarter upsells and renewals. At the same time, Snowflake CoCo is helping our engineers accelerate the migration of our core corporate data, converting legacy code and building pipelines in a fraction of the time it used to take. Together, these capabilities are helping us manage complexity and deliver better business outcomes, faster."

Docusign said the tools have changed how its teams work with enterprise data, especially for go-to-market staff who need account insights without writing SQL queries.

"CoCo and Snowflake's CoWork are transforming how Docusign operationalises enterprise data," said Bala Sahejpal, Chief Data Officer at Docusign.

"It has cut our data team's time to deliver value dramatically and also powers our conversational data agents. Our GTM teams can now carry out deep account research and get insights into an account without needing SQL expertise. Snowflake is an essential partner as we turn enterprise data into actionable intelligence for every team."

United Rentals said the products have reduced some work that once took days to minutes.

"As the world's largest equipment rental company, United Rentals is driving innovative uses of technology and data to strengthen our business and better serve our customers. Snowflake CoCo and CoWork are helping transform how our teams work," said Tony Leopold, Senior Vice President and Chief Technology and Strategy Officer at United Rentals.

"What once took days can now be done in minutes, enabling employees to use natural language to better understand their data, discover actionable insights, make more informed decisions and deliver greater value for our customers."

Governance focus

Another part of the announcement centres on governance and reuse. Users will be able to share workflows and plug-ins across organisations, set role-based profiles, and use memory functions that retain previous context and decisions while allowing teams to review or delete stored information.

Snowflake is also expanding how teams can publish and consume analysis. One employee's report or dashboard can be turned into a shared artefact inside CoWork, while CoCo can be used alongside it for further investigation or to build internal applications for more complex workflows.

Snowflake said it is bringing in the Mora team to strengthen these business intelligence-related features.

Alongside those updates, the company is launching an AI data package with more than 60 curated industry datasets, semantic views and a ready-to-use CoWork agent. It is also adding conversational discovery tools on Snowflake Marketplace so users can find, test and request data, applications and AI products through chat.

Model training

Snowflake used the update to outline broader work on AI model training and cost control. New model routing options for CoCo will allow teams to assign simpler background jobs to lighter models and reserve more advanced reasoning models for harder tasks.

The company also introduced Cortex Training, a service that lets organisations tailor open-weight models to their own data inside Snowflake's environment. Snowflake said the approach is intended to give customers more control over their models and lower training costs.

One early user is Poetiq, which said it is training a large language model for optimisation work on the platform.

"Poetiq is building the most powerful optimiser in the world using our own approach to recursive self-improvement. We have started to post-train the first LLM in the world that specialises in optimisation with Snowflake," said Ian Fischer, Co-Chief Executive Officer at Poetiq.

"Cortex Training gives us the ease of use and flexibility to handle everything we need to do from LoRA, to SFT, to full-parameter post-training with our custom GRPO implementation."