Factory's Droids Now Build and Migrate Software Inside Databricks
Factory's autonomous coding agents can now be purchased through the Databricks Marketplace and run directly against Unity Catalog governed data and models.
- Factory partnered with Databricks to run its Droids coding agents against Unity Catalog data and models.
- Factory is now sold on the Databricks Marketplace, with up to 10% of existing Databricks commitments applicable.
- Droids use frontier open and proprietary models served through Databricks Unity Gateway for inference.
- Three use cases: build data apps, migrate legacy platforms to Lakehouse/Lakebase, remediate Lakewatch security findings.
- Factory claims 9x faster feature shipping and 95% migration time reduction from early customers.
- Follows Factory's $200M Series raise at a $5B valuation and marketplace deals with Claude and OpenAI.
Factory has connected its autonomous coding agents, called Droids, to the Databricks Data + AI Platform. The integration lets the agents build, migrate, test, and secure software using the data, models, APIs, and governance controls that customers already manage in Databricks.
Governed models meet committed spend
The agreement has two commercial and technical components. Factory is available through the Marketplace listing, where eligible customers can apply up to 10% of an existing Databricks commitment toward Factory purchases.
Factory’s Droids can also use open-weight and proprietary models through Databricks Unity Gateway. Model selection and access therefore remain within Databricks’ governed model layer, alongside the customer’s existing permissions and policies.
Why data migrations need code changes
Factory describes Droids as agents that plan and execute multi-step engineering tasks, run tests, and open pull requests for human review. The company says its agents rank first on Terminal-Bench, a benchmark that evaluates performance on terminal-based software tasks. Production results will depend on each repository, test suite, permission model, and data environment.
The partnership grew from a migration for an unnamed global enterprise. Factory and Databricks are helping move a platform containing petabytes of data and hundreds of thousands of tables that support hundreds of business applications.
Application dependencies make that migration more complex than copying tables. Schema changes can break APIs, transformations, dashboards, and business logic, so the surrounding software must move with the data and produce equivalent results.
Three workloads, with one still planned
Factory groups the combined platform around three workloads:
- Build data and AI applications. Droids can turn product requirements into tested pull requests for Lakeflow pipelines, Spark jobs, Databricks Apps, and Agent Bricks agents governed through Unity Catalog. Factory says customers report shipping features nine times faster, though the announcement does not provide baselines or sample sizes.
- Migrate legacy data platforms. Droids can translate legacy warehouse and operational-store code for the Databricks Lakehouse and Lakebase, then validate the generated output for engineering review. Factory reports migration-time reductions above 95% for some customers.
- Remediate security findings. Factory can help teams remediate vulnerabilities in the applications monitored by Databricks Lakewatch. The proposed workflow would patch affected code, run tests, and open pull requests. This capability remains planned rather than generally available.
From Postgres to a tested pull request
Droids use Databricks APIs to create and run jobs, execute SQL validation queries, and deploy applications. A representative migration could follow this sequence:
- Inspect a legacy Postgres application, schema, and transformations.
- Create a Databricks Lakebase target and rewrite transformations in Python.
- Build and schedule a Lakeflow pipeline.
- Run SQL checks and compare outputs from the old and new systems.
- Create an AI/BI dashboard and submit the application changes for review.
That workflow brings database migration, application refactoring, validation, and deployment into one agent-run process. The connector’s credentials and Unity Catalog permissions determine which Databricks assets a Droid can access.
Distribution for Factory, automation for Databricks
Databricks gains application-code automation around its data services. Factory gains access to Databricks’ enterprise customers, a Marketplace procurement route, and an integration that can fit existing data-governance processes.
The companies already had financial and commercial ties. Databricks CEO Ali Ghodsi was an early Factory angel investor. Factory says it raised $200 million at a $5 billion valuation in September 2026, bringing total funding above $400 million. The company has also joined the Claude Marketplace and OpenAI’s business marketplace, allowing eligible customers to apply existing Anthropic or OpenAI spending commitments toward Factory.
What changes for Databricks customers
Existing Databricks customers can fund part of a Factory purchase from committed spend, route model access through Unity Gateway, and grant Droids scoped access to Unity Catalog assets. Factory’s setup guide describes a connector in Settings that enables Databricks-served models and managed MCP servers. MCP, or Model Context Protocol, is a standard that lets agents call approved tools and access connected data sources.
Claims to measure before rollout
Factory’s ninefold development-speed claim and 95% migration-time reduction lack published sample sizes, workload definitions, and measurement methods in the announcement. A useful evaluation would compare equivalent migrations, track engineer review time, measure validation failures, and include the work required to correct generated code.
The announcement also leaves several implementation details unspecified, including supported models and regions, data-retention terms, audit-log coverage, and pricing beyond the Marketplace commitment allowance. The Lakewatch remediation workflow has no published availability date.
Safe production deployment still depends on least-privilege service identities, protected branches, mandatory reviews, reproducible tests, and rollback procedures. Unity Catalog governs access to data assets, while repository and CI/CD controls govern the code that Droids generate.
Comments
