# Snowflake CoCo Adds Generally Available Snowpipe and Bulk File Loading

Snowflake CoCo can set up continuous Snowpipe ingestion or one-time COPY INTO loads. Review cloud access, Snowsight limits, approvals and costs.

By Paras Madan | 2026-10-28


Snowflake made continuous and bulk file loading through Cortex Code (CoCo) generally available on October 27. Developers can describe an ingestion job instead of assembling its setup manually, using CoCo CLI, Desktop or Snowsight. Both workflows support standard Snowflake tables and Snowflake-managed Iceberg tables. [GA announcement](https://docs.snowflake.com/en/release-notes/2026/other/2026-10-27-file-loading-coco-ga).

The first decision is whether files should load once or keep loading as they arrive. CoCo uses Snowpipe for continuous ingestion and `COPY INTO` for a one-time load. A historical export and an ongoing event feed need different plans, even if both live in the same bucket.

## Cloud setup has several paths

Continuous ingestion supports Amazon S3, Azure Blob Storage and Google Cloud Storage. CoCo can generate Terraform for the full pipeline, reuse a storage integration or prepare deployment files for a cloud team.

For S3, another route lets Snowflake configure the IAM role, bucket notifications and Snowflake objects after an approved temporary-access request, without collecting long-lived AWS keys. That option excludes AWS GovCloud and China deployments.

Temporary setup access does not mean the resulting pipeline's access expires. The setup permission lasts one hour by default, while the IAM role and bucket notification configuration remain for ongoing ingestion. [Continuous-ingestion setup guide](https://docs.snowflake.com/en/user-guide/data-load-snowpipe-auto-coco).

CLI and Desktop can apply deployment files with approval and local cloud credentials. Snowsight saves files for your team to download and deploy.

For continuous setup, CoCo shows SQL before changing objects. Default approval mode waits for permission, but bypass mode can run the displayed statements. Cloud authorization and deployment approvals still apply. Check the mode before relying on that review checkpoint.

## Bulk loading has different approval rules

A one-time load can use staged files, cloud storage or local files. CoCo selects format settings, creates or reuses a table and runs the load on a warehouse. Local uploads work through CLI or Desktop; in Snowsight, stage the files first.

The [bulk-loading guide](https://docs.snowflake.com/en/user-guide/data-load-bulk-coco) says CoCo waits for approval before object changes or loading data in every approval mode, including bypass mode. Do not assume the continuous-setup approval behavior applies here.

File interpretation still deserves review. XML and headerless CSV require supplied column definitions. CSV or JSON with multiline content needs the appropriate format setting; a mismatch can stop the load. Iceberg support is limited to Snowflake-managed tables, not external-catalog tables.

For an existing table, inspect the proposed column mapping and schema changes before approving. An assistant can reduce the SQL preparation, but the team still decides whether new input should alter the destination schema.

## Account for setup and ongoing ingestion separately

CoCo usage is token-billed, and inspection queries and bulk loads consume warehouse credits. Ongoing Snowpipe ingestion instead charges credits per GB loaded, without warehouse-time or per-file charges. Text formats are billed by uncompressed size; binary formats use stored size. [Snowpipe billing details](https://docs.snowflake.com/en/user-guide/data-load-snowpipe-billing).

Plan historical data explicitly, too. New-file notifications do not automatically backfill an existing bucket. CoCo can plan a recent-file refresh, but older files require a separate historical load. Agree on that boundary before accepting a pipeline as complete.

The useful change is a shorter path from an ingestion request to reviewable SQL and cloud configuration. Start with the desired load behavior, decide who can authorize infrastructure changes, and verify the resulting data. Those choices remain important even when CoCo prepares the commands.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders explain data-platform releases and AI workflows in clear content developers can read, use and share.

[Book a 30-minute call](https://cal.com/paras-madan/30min).