Snowflake's Data-Native Coding Agent
At Snowflake Summit 2026, held at the Moscone Center in San Francisco on 2 June 2026, Snowflake announced the relaunch of CoCo — formerly called Cortex Code — as a fully expanded enterprise coding agent. CoCo is not a general-purpose coding assistant pointed at a repository: it is an agent built to operate inside Snowflake's governed data platform, executing SQL queries, dbt models, and data pipelines against live, governed enterprise data. At launch, more than 7,100 Snowflake customers are already building with CoCo, and the company confirmed early deployments at Fanatics, Thomson Reuters, and WHOOP. The announcement also confirmed a native desktop application, multi-platform extensions, and an Anthropic partnership that places Claude models in Snowflake Cortex AI for production agentic workloads.
ADE-Bench: Why the Score Matters
The benchmark number that drew the most attention at the Summit was CoCo's 72.1 per cent pass rate on ADE-Bench — the Analytics Developer Evaluation framework created by dbt Labs to measure AI agents on real-world analytics and data engineering tasks. Claude Code running on Opus 4.7 and OpenAI Codex both scored 65.1 per cent on the same benchmark. Alongside the raw score, Snowflake published token efficiency data: CoCo uses 51 per cent fewer tokens than Claude Code on Opus 4.7 and completes tasks 8 per cent faster.
What ADE-Bench Actually Tests
ADE-Bench evaluates agents on tasks that closely resemble real data engineering work: writing SQL that correctly handles edge cases, building dbt models from a schema description, orchestrating multi-step pipelines, and diagnosing errors in existing data transformations. Unlike general coding benchmarks such as HumanEval or SWE-Bench, which test general software engineering on open-source repositories, ADE-Bench is specifically designed for the analytics and data engineering context. A score of 72.1 per cent on ADE-Bench is a more meaningful signal for a Snowflake customer than a higher score on a benchmark that primarily tests Python repository tasks.
Multi-Platform Access
CoCo launches with broader distribution than most enterprise coding agents. A native desktop application — CoCo Desktop — is available for download alongside a VS Code extension, an Excel extension, and an extension for Anthropic's Claude Code. A Slack bot and mobile application are in active development. New autonomous Cloud Agents capability allows teams to run CoCo tasks asynchronously without a developer actively supervising the session — starting an agent run, closing the laptop, and returning to a completed result. The breadth of access surfaces is relevant for data engineering teams whose work spans multiple tools in a single day: SQL in a browser, dbt in VS Code, analysis in Excel, and collaboration in Slack.
The Anthropic Partnership
Snowflake and Anthropic announced a deepened collaboration alongside the CoCo launch — Claude models now power Snowflake Cortex AI as the default backend for production agentic workloads. For enterprise teams that have already adopted Claude via the Anthropic API or Amazon Bedrock, the CoCo integration provides a governed data layer on top of the same model family they already use, reducing the context-switching cost of working across separate AI tools. The partnership also means teams building data-intensive agentic applications on Snowflake can access Claude's long-context capabilities without routing enterprise data outside their governed Snowflake environment.
What This Means for Indian Data Engineering Teams
India's data engineering talent base — concentrated in Bengaluru, Hyderabad, Pune, and Chennai — has been among the earliest adopters of cloud data warehousing, and Snowflake has a significant footprint across Indian product companies and IT services organisations. For teams running data engineering workloads on Snowflake, CoCo offers a direct path to accelerating dbt model development, SQL query authoring, and pipeline orchestration without routing enterprise data through external API endpoints.
The token efficiency advantage is particularly relevant in India, where enterprise AI budgets are typically more constrained than at US counterparts. Consuming 51 per cent fewer tokens than Claude Code on equivalent analytics tasks translates directly into lower API billing costs — a meaningful difference for teams running high-frequency query generation or automated reporting pipelines at scale. Data teams evaluating AI-assisted development tooling should include CoCo in their pilot evaluation alongside general-purpose coding agents, since the ADE-Bench gap suggests data-native specialisation delivers measurable productivity benefits on the task categories that most frequently occupy data engineers.
The Bottom Line
Snowflake CoCo, announced at Snowflake Summit 2026 on 2 June 2026, scored 72.1 per cent on dbt Labs' ADE-Bench analytics evaluation — outperforming Claude Code and OpenAI Codex at 65.1 per cent each, while using 51 per cent fewer tokens. The agent runs natively inside Snowflake's governed data platform, powered by Anthropic's Claude via Cortex AI, and is now available across a native desktop app, VS Code, Excel, Slack, and the Claude Code extension. For Indian data engineering teams on Snowflake, CoCo's data-native advantage, measured performance edge, and token efficiency make it the first enterprise coding agent specifically worth evaluating for analytics and pipeline work ahead of general-purpose alternatives.
Frequently Asked Questions
What is Snowflake CoCo and how is it different from other AI coding agents?+
Snowflake CoCo, announced at Snowflake Summit on 2 June 2026, is an enterprise coding agent formerly known as Cortex Code. Unlike general-purpose coding agents such as GitHub Copilot, Cursor, or Amazon Kiro — which operate on arbitrary codebases — CoCo is purpose-built to operate inside Snowflake's governed data platform. It executes SQL queries, dbt models, and data pipelines directly against live enterprise data, giving it contextual access to schemas, governance policies, and data lineage information that general coding agents cannot access. This data-native design is the source of its performance advantage on analytics-specific benchmarks.
What is ADE-Bench and what does CoCo's score of 72.1% mean?+
ADE-Bench is the Analytics Developer Evaluation framework, created by dbt Labs to measure AI agent performance on real-world analytics and data engineering tasks — writing SQL that handles edge cases correctly, building dbt models from schema descriptions, orchestrating multi-step pipelines, and diagnosing errors in data transformations. CoCo scored 72.1 per cent on ADE-Bench at Snowflake Summit 2026, compared to 65.1 per cent for both Anthropic Claude Code on Opus 4.7 and OpenAI Codex. CoCo also uses 51 per cent fewer tokens than Claude Code and completes tasks 8 per cent faster on equivalent ADE-Bench tasks.
Where can developers access Snowflake CoCo and what platforms does it support?+
Snowflake CoCo is available to the more than 7,100 Snowflake customers currently building with it. Platform access includes a native desktop application called CoCo Desktop, a VS Code extension, an Excel extension, and an extension for Anthropic's Claude Code. A Slack bot and mobile application are in active development. Autonomous Cloud Agents allow CoCo to run tasks asynchronously without an active developer session — teams can start a data engineering job and return to the completed result without keeping a laptop running. CoCo is powered by Anthropic's Claude models through Snowflake Cortex AI.
Which Indian data engineering teams should evaluate Snowflake CoCo?+
Indian data engineering teams running SQL, dbt, and pipeline workloads on Snowflake are the most direct audience for CoCo. The token efficiency advantage — 51 per cent fewer tokens than Claude Code on analytics tasks — translates to lower API billing costs, which is materially relevant for teams with constrained AI budgets running high-frequency query generation or automated reporting pipelines. Enterprise teams in banking, financial services, e-commerce, and media — sectors with significant Snowflake analytics workloads — should include CoCo in their AI tooling evaluations, since the ADE-Bench performance gap suggests data-native specialisation delivers measurable productivity benefits.
Written by
TechPillow Team
Sharing insights on technology, product development, and the Indian tech ecosystem.