Three stories, all from this week, all worth your time.


1. Hermes Agent v0.18.0 "The Judgement Release" Arrives With 2,500+ Likes

Nous Research dropped what the community is already calling the most significant Hermes Agent update to date: v0.18.0, officially tagged "The Judgement Release." The headline feature is native Mixture-of-Agents (MoA) support — multiple LLMs collaborate on the same task simultaneously, cross-checking each other before a final answer emerges. Alongside MoA, the release ships evidence-based task verification via /goal contracts (the agent verifies completion using actual tests and evidence, not just confidence), a /learn command that instantly converts any directory, URL, or recent chat into a reusable skill, and background subagents that persist without blocking the main thread. The sheer scale of the effort is staggering: 1,700+ commits, 1,000 pull requests merged, 370+ contributors, and 900+ issues closed — all within a 12-day sprint. The official @NousResearch announcement pulled over 2,500 likes and sparked a wave of threads from builders sharing early results. The running community consensus: MoA isn't just an incremental improvement — it's the feature that finally makes Hermes feel like it's reasoning, not just executing.

https://x.com/NousResearch/status/2072413332665962617


2. @JulianGoldieSEO's Builder-Judge Loop: Make Your Agent Rewrite Itself Until It's Good Enough

Julian Goldie shared what may be the most practical Hermes Agent technique making rounds on X right now: the builder-judge loop, also circulating as the "Ralph loop." The setup is elegant in its simplicity — a "builder" model drafts the output, then a separate "judge" model scores it on a defined scale (a live example from Goldie's thread: 54/100 on the first pass), identifies specific weaknesses, and sends it back for revision. The loop repeats until the judge issues a passing score — in the same example, 92/100 by the third iteration, with no human needed at any checkpoint. Goldie walked through live demonstrations from SEO content pipelines, code review cycles, and research report generation, all showing meaningful quality lifts without touching the underlying model or changing the cost structure. The core reframe is what landed hardest with the community: "You don't need a better model. You need a better loop."

https://x.com/JulianGoldieSEO/status/2072258798404903141


3. @amandaorson's Hermes Agent as a $5/Month AI Chief of Staff

Amanda Orson, Founder and CEO of Galleon (real estate tech), shared the most grounded business case for Hermes Agent to make the rounds this week: she runs it as her full-time AI Chief of Staff across Slack and Linear for less than $5/month in compute. The setup handles calendar management, daily briefings, task triage, and follow-ups — and critically, it remembers context across weeks without needing to be re-briefed each session. For project management, she pipes standups and action items from sales calls directly into Linear tickets, with Hermes flagging when things stall and keeping roadmaps moving. Her own benchmark: "roughly 70% as effective as a human assistant at a fraction of the cost." The thread resonated because it's not a developer showcase or a wild automation story — it's a real operator describing a real workflow that saves her real time, and the sub-$5/month figure makes it impossible to dismiss as enterprise-only.

https://x.com/amandaorson/status/2066927382783504598


Spotted something great on X about OpenClaw or Hermes Agent? Share it — the community is the best part.