Dev Builds Fully Agentic AI Newsroom on MiniMax M3, Burning 1-2B Tokens Daily
robleclerc · x · 2026-07-25
Developer Rob Leclerc shared his fully agentic newsroom project, which burns 1-2 billion tokens per day and runs entirely on the MiniMax M3 model.
He detailed the workflow: for every step—from sourcing, selection, research, angle generation, drafting, fact-checking, and editing, to title and lede generation—he establishes a baseline using the latest frontier models. He then iterates on the harness until M3 meets or exceeds the baseline. The process has never failed so far.
More from coding & agent
- Opus 5 works well in simple coding, but breaks an autonomous Compound Engineering flow — danshipper · 2026-07-25
- Claude Opus 5 lands in Cursor, matches Fable 5 on CursorBench at half the price — dean_rie · 2026-07-25
- Running coding agents on two machines exposed the real problem: the control plane — Competitive-Bend-143 · 2026-07-25
- Agent memory quality depends more on write gating and dedup than on retrieval — chirayusir · 2026-07-25
- KG-RAG demo models 53K rights records and runs locally with auto-repair — JeremyCMorgan · 2026-07-25
- LangChain hosts a production-focused chat on building an AI SRE — LangChain · 2026-07-25