By Hamza Faisal
Factory has released Droid 1.0, the largest update yet to its coding agent harness, and its headline change is a different way of thinking about how AI agents use tools. Instead of a model firing off one tool call, waiting for the result, and then firing the next, Droid 1.0 lets the model write a JavaScript program that runs inside a sandbox and executes hundreds of tool calls in a single pass.
The company calls this programmatic tool calling, or Script mode, and it is now the default in Droid. Factory defines a harness as everything around the model: the prompts, the tools, how results come back, and the loop that runs it all. Droid 1.0 is a model-independent harness, meaning it works with whichever model a developer plugs in, including through their own API keys.
One program instead of hundreds of round trips
The practical payoff, according to Factory, is fewer wasted model calls. A batch of unit tests that previously took 10 model calls now takes 5, the company says. In one internal example, 151 issues were moved and verified in 312 tool calls using just 2 model requests and roughly 61,000 tokens.
The leaner design shows up in the prompt itself. Droid 1.0's base prompt shrank 37 percent, from 20,887 input tokens to 13,055, and the number of tool definitions shipped up front fell from 11 to 2. Because previously processed prompts stay cached when new tools are discovered, Factory says a typical production request now reads 98 percent of its input from the prompt cache. The company also reports that Script mode cut costs by 15 percent with Claude Opus 5.5 compared with direct tool calling.
Benchmark claims, with a caveat
On Terminal-Bench 4, Factory's self-reported numbers put Droid at a 65.2 percent pass rate using Claude Opus 5.5, at $4.73 per passed task. The company says that beats Cursor, OpenCode, Devin and Claude Code on the same model by 3.0 to 7.6 points at up to 24 percent lower cost, and that with GPT-6.1 Sol it scored 3.5 points higher than Codex at 12 percent lower cost. Those figures come from Factory's own runs between October 23 and 28, so they should be read as vendor claims rather than independent results.
The approach builds on CodeAct, a 2024 research idea in which agents act by writing code rather than by emitting individual tool calls. Factory says its scripts run in a sandbox without direct file, network or system access, a guardrail that matters when an agent is executing generated programs against a real codebase.
Available now, from desktop to SDK
Droid 1.0 is available now as a desktop app for macOS and Windows, as a terminal interface, and through a Node SDK called @factory/droid-sdk that runs the same harness as the command-line tool. Paid plans start at $20 per month.
Behind the release sits what Factory calls the Agent Factory, internal automation that tracks more than 20 models, adds each new model and snapshot to Droid, and checks it against behavioral assertions that serve as a minimum bar for release. For a market crowded with coding agents that all claim to be faster, Droid 1.0's pitch is narrower: the model is not the differentiator anymore, the harness around it is.



