| Hitting a million brick walls with multi-turn RL training isn't fun, so I thought I would try something new to climb Stanford's leaderboard for now! So this weekend I was just tinkering with multi-agent systems and... somehow ended up beating Claude Code on Stanford's TerminalBench leaderboard (#12)! Genuinely didn't expect this - started as a fun experiment and ended up with something that works surprisingly well. *What I did:* - Built a multi-agent AI system with three specialised agents: - Orchestrator: The brain - never touches code, just delegates and coordinates - Explorer agents: Read & run only investigators that gather intel - Coder agents: The ones who actually implement stuff - Created a "Context Store" which can be thought of as persistent memory that lets agents share their discoveries. - Tested on TerminalBench with both Claude Sonnet-4 and Qwen3-Coder-480B. *Key results:* - Orchestrator + Sonnet-4: 36.0% success rate (#12 on leaderboard, ahead of Claude Code!) - Orchestrator + Qwen-3-Coder: 19.25% success rate - Sonnet-4 consumed 93.2M tokens vs Qwen's 14.7M tokens to compete all tasks! - The orchestrator's explicit task delegation + intelligent context sharing between subagents seems to be the secret sauce *(Kind of) Technical details:* - The orchestrator can't read/write code directly - this forces proper delegation patterns and strategic planning - Each agent gets precise instructions about what "knowledge artifacts" to return, these artifacts are then stored, and can be provided to future subagents upon launch. - Adaptive trust calibration: simple tasks = high autonomy, complex tasks = iterative decomposition - Each agent has its own set of tools it can use. *More details:* My Github repo has all the code, system messages, and way more technical details if you're interested! (Github handle is danau5tin). Thanks for reading! Dan |