Factory has released Droid 1.0, a model-independent coding agent harness featuring programmatic tool calling (PTC) as the default interaction mode. Instead of returning every tool result to the model context, PTC employs a Script tool where the model writes JavaScript executed in a sandbox; only printed output returns, allowing one request to drive hundreds of concurrent tool calls. This reduces the base prompt 37% to 13,055 tokens and cuts initial tool definitions from 11 to 2, with rarely used tools loaded on demand. On Terminal-Bench 4, Droid 1.0 achieved a 65.2% pass rate with Claude Opus 5.5 at $4.73 per passed task—outperforming Cursor (62.1%), OpenCode (62.1%), Devin (59.6%), and Claude Code (57.6%)—while reducing cost up to 24% and time per passed task 16–26%. With GPT-6.1 Sol, Droid scored 57.6% at $0.79 per task versus Codex (54.0%) and OpenCode (51.0%), though Codex remained 10% faster per passed task. Internal benchmarks show a unit-test fix completed in 5 model calls (41.6s) versus 10 (56.8s), and a CI sweep of 5,384 jobs executed in 2 requests with under 0.1% of tool results reaching context. Additional harness improvements include a prompt cache serving 98% of input tokens, bounded autonomy triggers at 3 and 6 identical calls, compaction summaries doubled to 16,000 tokens, and desktop app first-paint latency reduced from 1,839 ms to 488 ms. The SDK (@factory/droid-sdk) is Apache 2.0 licensed; paid plans start at $20/month.