Factory Acquires Typedef to Give Its Droids Agents Smarter Enterprise Code Awareness

Factory has acquired Typedef, a startup that builds dependency maps for enterprise data systems, according to the company’s acquisition announcement . Factory plans to integrate Typedef’s context layer into Droids, its coding agents for planning and executing software tasks inside customer environments. The company did not disclose the purchase price, whether it paid in cash or stock, an integration schedule, or a product release date.

Typedef analyzes code to identify relationships among data models, pipelines, services, and dashboards. Its dependency graph could give Droids evidence about which systems may be affected when an agent changes a column, renames a table, or refactors a transformation.

A dependency map for risky changes

Large enterprise data systems often contain dependencies that exist in code but remain undocumented elsewhere. A field defined in one model may feed several transformations, dashboards, and customer-facing services. Changing it without tracing those connections can cause failed jobs or, more dangerously, results that remain plausible while becoming incorrect.

Data lineage records how information moves and changes across a system. By extracting that lineage from code, Typedef aims to help agents assess the downstream effects of a proposed change before editing files or running a migration.

Target workflows

• Migration planning: Identify coupled models, services, and reporting systems before changing schemas.
• Pipeline debugging: Trace incorrect or missing data back through upstream transformations.
• Model refactoring: Find downstream consumers and validate that a change preserves expected behavior.

Factory describes Droids as the execution layer that operates within a customer’s environment, while Typedef supplies structured context about the data system. The combined technology is intended to help both Droids and Factory engineers plan changes, gather supporting evidence, and validate results.

The announcement does not identify which warehouses, transformation frameworks, orchestration tools, or business intelligence platforms will be supported. It also leaves open whether customers will access the dependency graph directly or only through Droid workflows.

Two repeat founders join Factory

Typedef co-founder Yoni Michael previously co-founded Coolan, which applied machine learning and infrastructure analytics to data-center reliability and efficiency. Coolan’s technology ran across more than 100,000 servers before Salesforce acquired the company in 2016. Michael subsequently led infrastructure engineering teams at Salesforce and Tecton.

Co-founder Kostas Pardalis previously founded Blendo, a data integration company used by more than 400 businesses. He later held product leadership roles at RudderStack and Starburst, both of which build infrastructure for moving, managing, or querying enterprise data.

Capital accelerates Factory’s acquisition strategy

Factory has used acquisitions to add teams with specialized agent infrastructure. Typedef follows its purchase of Lumetric, which was developing model-agnostic AI workflows for the next generation of Factory Desktop.

Factory reported in April 2026 that hundreds of thousands of developers used Droids daily, including developers at NVIDIA, Adobe, EY, Palo Alto Networks, and Adyen. Organizations of that size commonly operate extensive warehouses, legacy pipelines, and reporting systems where incomplete lineage increases the risk of automated changes.

The harness gains deeper context

Software-engineering agents combine a language model with a harness that assembles context, exposes tools, manages execution, and enforces permissions. Factory’s strategy emphasizes that surrounding system because the same model can produce different results depending on the files, tools, feedback, and constraints available to it.

Factory said its Droid scored 58.75% with Claude Opus 4.1 on Terminal-Bench and ranked first when the company reported the result. Terminal-Bench evaluates whether an agent can complete command-line tasks in a terminal environment, while SWE-bench measures performance on software issues drawn from real repositories. Rankings can change as benchmarks, agent configurations, and models are updated.

Lumetric added workflow technology around the agent, and Typedef adds structured knowledge about enterprise data dependencies. Together, the acquisitions broaden Factory’s harness from code execution toward system-aware planning and validation.

Factory also says customers should retain control of their infrastructure and data. Its product positioning casts engineers as designers and operators of AI-assisted software systems, with Droids performing delegated work inside defined environments.

What teams can measure

Factory has not published evidence showing how Typedef changes Droid performance on these workflows. Once the integration ships, evaluations can compare dependency coverage, task completion rates, incorrect edits, review time, and recovery from failed changes.

Questions the announcement leaves open

• Which technologies will Typedef support, including SQL warehouses, dbt -style projects, orchestrators, and business intelligence tools?
• How will the graph handle dynamic SQL, generated schemas, runtime transformations, and dependencies outside source control?
• Will lineage update incrementally as code changes, and how quickly will agents receive refreshed context?
• Where will metadata be processed and stored, and what data-retention controls will customers receive?
• How will Droid permissions and human approval requirements apply to migrations and production changes?
• When will the integration become available, and will Factory charge separately for it?

Factory is betting on system awareness

Typedef gives Factory a specialized team and technology for mapping the systems around enterprise code. If the integration works as described, Droids should be able to plan changes with clearer evidence about downstream effects and validate more than the edited files alone. The practical result will be measurable in fewer missed dependencies, faster diagnosis, and safer migrations.