Jive rebuilds the agentic loop with DAG 'graph calls', beating Codex 7x on speed
merijjeyn · reddit · 2026-09-25
The author argues the current LLM-call → tool-call agentic loop is outdated, and rebuilds it around System 1 / System 2 thinking: use full reasoning power to plan an execution graph, then execute the plan with cheap "fast thinking".
Approach: Jive replaces tool calls with "graph calls" — each graph is a DAG of bash nodes and jev nodes that can declare dependencies and reference each other's outputs. The model lays out its own execution flow while reasoning, then walks through it with jev calls, eliminating redundant LLM calls.
Benchmarks: Across 6 tasks vs Codex and Claude Code, Jive averages 2m31s / 8.7k tokens, versus 17m40s / 16.2k for Codex and 12m12s / 32.7k for Claude Code — fastest and cheapest on every task. It shines at repo investigation, bulk classification, multi-step profiling, repetitive edits, and eval workflows.
Caveats: Accuracy on a large-scale SWE benchmark is not yet verified; the author reports never seeing it underperform other harnesses in personal use, and has open-sourced the code inviting discussion and contributions.
Related event: Open-source Jive Rebuilds Agent Loop with Fast/Slow Thinking(2 posts)→
More from coding & agent
- Pigment Agents now reads and posts in Slack, inheriting user permissions — mattturck · 2026-11-12
- Map of the agent economy: 50+ startups across payments, spend control, identity, memory — venturetwins · 2026-11-12
- OpenCode launches Browse, an on-device browser agent with local logins — Saul_Loveman · 2026-11-12
- VentureTwins maps the B2A economy: dozens of agent infra and app companies in first series installment — venturetwins · 2026-11-12
- Google releases 76-page 'Agents Companion' whitepaper on agent engineering — mdancho84 · 2026-11-12
- Spinning up containers for 20-line agent tools is backwards: Safescript compiles provable, 50ms-deploy apps — uriwa · 2026-11-12