Factory Droid 1.0 Makes Its Coding Agent Write Scripts Instead of Single Tool Calls

Factory has released Droid 1.0, a new version of the engine that runs its software development agent. The company says the update makes the agent more accurate, more efficient and faster on everything from large projects to everyday coding work.

Writing scripts instead of single calls

The biggest change is that programmatic tool calling is now on by default. Normally, a coding agent calls one tool, waits for the answer and then decides what to do next. Droid 1.0 instead writes a small program that can run up to hundreds of tool calls in a single model request.

Factory gave a practical example. Moving 151 Linear issues and checking each one used to take 312 separate tool calls. Done as direct calls, that meant about 36 requests and 3.1 million input tokens. Run as a script, the same job took just two requests.

Benchmarks and testing

On Terminal-Bench 4, Factory says Droid beats the native harnesses that popular models ship with, with up to 20% better efficiency. The team also describes an internal system it calls Agent Factory, which tests every change against every model family it supports. That matters because Droid is meant to work across several models, not just one.

Why it caught my eye

I don't usually write about developer tools, but plenty of people building fan sites, bots and small animation pipelines rely on these agents now. Batching dull chores like bulk ticket moves or file cleanup into one script is a sensible idea, and it should cut both waiting time and cost. I'd still treat vendor benchmarks as a starting point and check it on your own repos, since scripted runs can also fail in bigger ways than single steps.

Source: Factory's Droid 1.0 announcement on X.