Paper: Code-First Tool Calling Beats JSON for Agents
eyishazyer · x · 2026-08-15
A perspective suggests JSON tool calling is an unnecessary bottleneck for newer AI agents, and capable models can orchestrate tools better by writing code.
A paper compares standard JSON calls with a code-first setup where the model writes one Python script to invoke the same tools. Across 14 models on a 309-task BFCL v4 subset, programmatic tool calling matched or beat JSON in 11 models.
Related event: Study: Code-First Tool Calling Outperforms JSON(2 posts)→
More from coding & agent
- OpenAI Web App GitHub Connector Bypasses Usage Limits? — daniel_mac8 · 2026-08-15
- Enterprise Agent Skills Need Clear Ownership: Public Data Reveals Management Pain Points — myfear3 · 2026-08-15
- Developers Report Claude Code Hitting 5-Hour Usage Limit Too Fast — AccordingLeague9797 · 2026-08-15
- CMU PhD: Agents should empower, not just automate — stanfordnlp · 2026-08-15
- 6 Core Questions for Building Your Agent Workflow — brandon_galang · 2026-08-15
- Continue: AI coding assistant integrated into your IDE — ZabihullahAtal · 2026-08-15