I made Claude compete against itself. The smartest model does not automatically make the best agent. To prove it, I took the same Opus 4.6 model, initial prompt, and empty repo and but it inside two different coding harnesses: Claude Code vs. Factory Droid. The task: clone Excalidraw from... scratch, inspect the original in a browser, implement its key interactions, and verify the result. Factory Droid: 8 minutes, 20 tool calls, $1.60 Claude Code: 19 minutes, 40 tool calls, $1.87 The only difference was the harness. Get Free Factory credits here:show more

@FactoryAI So many activities to do with those 11 minutes saved!

🫡

yes sir chief

@FactoryAI the software factory is in good hands 👀

@FactoryAI was great to collaborate on this 🤝

@FactoryAI 🔥

@FactoryAI Factory harness is one of the best, my most used daily driver for a reason

@FactoryAI what model is your goto?

@FactoryAI GLM 5.2 (can't wait for 5.3 in factory!!) and deepseek v4 flash, both being droid core means i get to use my 5 hour limit and then my 5 hour droid core limit so i have longer sessions.

@FactoryAI droid is gud ... very gud

@FactoryAI very gud

@FactoryAI 同一个脑子换套工具链，执行力差距可能比模型差距还大。

@FactoryAI very cool

@FactoryAI u ran a twin study and nurture won

@FactoryAI Why use Opus 4.6 instead of 5?

@FactoryAI better highlights the power of the harness versus just the model

@FactoryAI Because they have made the harness part of the model more over time?

@FactoryAI using the same 4.6 model across both makes the difference in performance much more attributable to the harness.

@FactoryAI Right, could but you not have used Opus 5 on both?

@FactoryAI Droid used to be my daily driver and it was really good. Maybe this is my sign to start using it again 😅

@FactoryAI same Opus

@FactoryAI can droid use your claud subs or codex subs or grok subs

@FactoryAI fewer tool calls might just mean it checked its own work less. did you diff the two clones side by side, or is this speed and cost only?

@FactoryAI yup, the both had a criteria of functionality they had to pass (and both passed) and you can see the actual output and code that are very identical

@FactoryAI that's a solid comparison

@FactoryAI Completion gets a twist in our harness as tasks have deadlines. Meeting briefs wait on a research agent for 45 min but ship partial 10 min before meeting start. Partial brief before the meeting beats complete one after. How does the droid loop define done when the clock decides?

@FactoryAI What are you currently working on?