Snowflake Debuts CoCo and Datastream for Developers

Snowflake announced major new capabilities for CoCo and a new streaming service, Snowflake Datastream, at Snowflake Summit 2026 on June 2, per Snowflake's press release. CoCo (formerly Cortex Code) is positioned as a coding agent inside what Snowflake calls its "agentic control plane," giving builders a governed environment to work across data, models, and apps. Datastream is described as a fully managed, fully Kafka-compatible streaming service that writes real-time data directly into Snowflake tables. Christian Kleinerman, Snowflake's EVP of Product, said CoCo "makes work for experienced developers dramatically faster and easier, and is opening the door for non-traditional builders, like analysts and data-savvy business users, to start creating pipelines, automations, and AI apps on their own," per the press release.

According to Snowflake, CoCo is now available as a native desktop app (CoCo Desktop) and also reaches builders through a mobile app, a Slack bot, and extensions for VS Code, Microsoft Excel, and Anthropic's Claude Code. Snowflake said CoCo can run tasks autonomously through new Cloud Agents that execute in the cloud rather than on a local machine, supported by role-based access controls, audit trails, and a secured local sandbox. Snowflake named Fanatics, Thomson Reuters, and WHOOP as early customers.

Independent coverage from SiliconANGLE and Techzine reports Snowflake's claim that CoCo achieved a 72.1% pass rate on ADE-Bench, ahead of Anthropic's Claude Code and OpenAI's Codex at 65.1%; ADE-Bench is a framework created by dbt for evaluating agents on analytics and data-engineering tasks. Aimpoint Digital additionally cites a 51% fewer tokens efficiency figure. Editorial analysis: vendor-run benchmarks are one signal but typically use task sets and settings that favor the sponsor's integration, so practitioners should validate accuracy, latency, and token cost on their own workloads before treating these results as equivalence to general-purpose coding agents.

Snowflake frames Datastream as a way to avoid running a separate streaming system alongside the data platform: organizations get standard Kafka compatibility for ingestion plus Snowflake-native governance, masking, and lineage, with Snowflake citing a $128 billion market opportunity. Editorial analysis - industry context: a managed, Kafka-compatible layer can reduce the connector and cluster maintenance that often burdens real-time pipelines, but the questions that decide production fit are ingestion latency, exactly-once semantics, schema evolution, and cost.

Public coverage frames the announcements as part of a broader "agentic enterprise" push, where platforms supply orchestration, governance, and shared context rather than only model endpoints. Diginomica reported that CEO Sridhar Ramaswamy used the opening keynote to argue that agents are reshaping how work gets done and to emphasize a control plane that coordinates multiple agents and context layers. For teams running Kafka-to-warehouse pipelines, a native managed streaming service could shrink integration surface area, while embedding a coding agent across desktop, Slack, and IDEs aims to shorten iteration loops.

Watch for early Datastream users to report ingestion latency, exactly-once behavior, and schema-evolution handling, and for whether Snowflake or third parties publish reproducible benchmarks for CoCo's accuracy and token-efficiency claims. Governance and observability for agent-initiated data changes, including lineage and audit trails, will be a key operational test as Snowflake positions CoCo inside a control plane.