This was a week of practical releases rather than big launches. Snowflake added a feature its users have asked about for years, Databricks removed a setup step that stopped many teams from connecting AI assistants to their data, and both platforms made OpenAI's new agent model available. Here is the short list from 21 to 27 October 2026 for analysts, analytics engineers and data engineers. Last week's edition covers 13 to 20 October.
Snowflake
Snowflake, 21 to 27 October
Declarative Feature Store (GA)
Define entities, feature views and groups in YAML or Python and deploy them with a plan-and-apply CLI.
Pipeline Builder (preview)
A visual designer that wires notebooks, Python, ML Jobs and SQL into task graphs kept in Git; skill and plugin sharing in CoCo goes GA.
Responses API and Cortex Search from SQL (preview)
An OpenAI-compatible Responses API endpoint for OpenAI models, plus a table function for low-latency Cortex Search queries.
Secondary indexes (GA)
Indexes on standard tables for selective lookups and ranges, billed as search optimization.
File loading with CoCo (GA)Now
Describe the files and CoCo builds a continuous Snowpipe pipeline or a one-time load, including Terraform.
Secondary indexes, in one minute
Until now, standard Snowflake tables had no indexes: speed came from micro-partition pruning, clustering and search optimization. A secondary index maps column values to the micro-partitions that hold them, so a query such as all orders for customer 42 reads only a few partitions. With INCLUDE, the index can return the answer without touching the table. It needs Enterprise Edition and does not work on Iceberg tables or on columns with masking policies. Read the full guide: Snowflake secondary indexes explained.
CREATE INDEX idx_orders_customer
ON orders (customer_id, order_date)
INCLUDE (status, amount);- CoCo file loading
- Snowflake can create the AWS IAM role and bucket notifications for S3 after one temporary approval, so no long-lived access keys are needed. It can also generate Terraform for any cloud.
- Pipeline Builder
- Drag notebooks, Python files and SQL files onto a canvas; connections set the run order. Each pipeline is a project you can review and keep in Git.
- Ingress policies (preview)
- One allow-or-block rule that combines network, authentication, client, location and endpoint conditions.
- Task graphs (preview)
- Update pipelines without suspending them, and run tasks depending on whether earlier ones succeeded.
Databricks
AI assistants and agents
21 to 27 October- ChatGPT, Codex, Claude, Claude Code and Cursor can connect to Databricks MCP servers with default OAuth apps (GA, 27 October)
- Genie Code tasks in Lakeflow Jobs (GA): run a prompt on its own as a scheduled job step
- Unity Gateway screens MCP tool calls and results for jailbreak and prompt-injection attempts (beta)
Connecting an AI assistant to governed data no longer needs an admin ticket.
Compute and developer experience
22 to 27 October- DataFrame caching with df.cache() on serverless compute (preview)
- Serverless jobs can set their Python environment from pyproject.toml (beta)
- Edit files in Unity Catalog volumes in the workspace editor (GA)
Fewer workarounds when moving notebooks to serverless.
Connectors
beta- QuickBooks, PayPal, Google Analytics (aggregated GA4 reports), Microsoft Entra ID and Box audit logs
- Lakeflow pipeline system tables for monitoring updates (GA)
More sources without writing your own ingestion code.
OpenAI's GPT-6.1 Terra reaches the data platforms
OpenAI released GPT-6.1 Terra this week, a model built for fast, high-volume agent work such as using software through its interface. It costs $1 per million input tokens and $5 per million output tokens, half the price of GPT-6.1 Sol, and OpenAI reports a slightly higher score than Sol on its Agents' Last Exam benchmark at about half the cost per task. Databricks added it to Unity Gateway on 27 October, and Snowflake's new Responses API endpoint lets clients call OpenAI models through Cortex. Read GPT-6.1 Terra explained for the benchmarks and safety findings, and the best AI models in October 2026 to see how it compares.
Coming up: Snowflake Expedition, 3 to 5 November
Snowflake's new flagship virtual event replaces BUILD. It is free, and each day targets a different audience: technical leaders and builders, then sales and marketing, then finance. Snowflake says it will show new ways to connect data from any source, ground AI in business context and move agents into production. We will cover the announcements in next week's edition.
Check your understanding
1/3Which feature lets a Snowflake query skip most of a table when looking up one customer?
Secondary indexes map values to the micro-partitions that contain them.
2/3What did Databricks make GA on 27 October for ChatGPT, Claude and Cursor?
Users connect with their own permissions, without an admin creating an OAuth app first.
3/3How does GPT-6.1 Terra's token price compare with GPT-6.1 Sol?
Terra costs $1 input and $5 output per million tokens, against $2 and $10 for Sol.
You got 0 out of 3.
More platform news: last week's data platform news, Snowflake October 2026 update, Databricks October 2026 update and Snowflake vs Databricks for AI agents.
Sources: Snowflake release notes (20 to 28 October 2026) and the secondary indexes, CoCo file loading and Pipeline Builder release notes; Databricks platform release notes, October 2026; InfoWorld on GPT-6.1 Terra (27 October 2026); Snowflake's Expedition 2026 announcement. Features roll out by region and edition; check availability in your account.
Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 28 October 2026. About me