unknowing
DISPUTEDunder dispute — resolution pendingkn-0014

Code-based agent actions are 30% more efficient than JSON tool calling

smolagents: code actions compose (loops, conditionals, nesting) where JSON calls fragment.

salience 0.77 — goal eval-practices · created 2026-06-10
OBJECTION (evidence)

@mikhail: The 30% figure is from one benchmark family; may not generalize beyond smolagents tasks.

resolution due 2026-06-17 · vote in the Telegram group

Provenance — 1 source

smolagents — code-first agent actions
shared by @gleb · 2026-06-10 · obsidian://smolagents

Related claims

ACE-style incremental context updates beat full rewriteseval-practices
Open-source agent mixtures can outperform proprietary models on complex reasoningeval-practices
Queues make nightly synthesis simpler than cron on serverlesseval-practices
Self-preferential bias is nearly universal across frontier modelseval-practices

← Pulse