Field guide / primary domain
25
sources in this field
Updated August 13, 2026
Current thesis
The shortest path to orientation.
Hermes (Nous Research) is a full agent OS with discovery, distribution, coordination, maintenance. Single Codex CLI/GPT-5.5 backend runs 24/7 ops for $100/month. Hermes Workspace consolidates chat, memory, skills, terminal, files; v0.12.0 added Kanban coordination replacing multi-terminal chaos. Curator weekly consolidates skills by usage; Atlas is a 100+ tool directory with live GitHub data. Integration maturity gates adoption: Google Workspace first (required for workflow), Firecrawl for search, Browserbase for automation, Composio (hours→minutes integration). Template recipe—personal agent + workspace UI + curator + discovery + task board—converges ecosystems. Gumclaw (Gumroad's business agent) runs Fable 5 on cron-woken Mac, storing all persistence in filesystem (policies, logs, people notes, ledgers, indexed repos). Each session reads permanent institutional memory; incidents become dated policy rules every session enforces, turning error corrections into durable knowledge without fine-tuning. Email and LinkedIn approvals require separate switches; editing any message re-locks approval. Hard rule: prospect replies must pause all outreach channels before classification—never auto-halt based on model judgment. Autonomy escalates one layer at a time: automate research, then monitoring, then low-risk actions, gating high-risk/external behind explicit approval until preceding layer proves reliable.
Evidence board
Claims worth carrying forward01Core GTM agent pipeline pattern: market -> research -> signal -> message -> approval -> action -> CRM. The workflow structure matters more than which specific tools (Apollo, Clay, Crustdata, HubSpot, Smartlead, Unipile) fill each slot.
02ICP definition should specify four things upfront: target companies, target buyers, explicit exclusions (existing customers, active opportunities, competitors, suppressed contacts), and minimum required fields before outreach (domain, named buyer, current signal, source, retrieval date).
03Use a 100-point account scoring rubric: company fit 25, observable problem 20, recent buying signal 20, correct buyer 15, offer relevance 10, evidence quality 10 — every point requires a cited, dated reason so the agent can say 'we do not know' instead of inventing pain.
04Qualified accounts should not go straight to outreach — they enter a monitoring state waiting for a genuine timing signal (funding round, exec hire, product launch, hiring, tech change, compliance change) with event/date/relevance/source/expiry fields, since old signals go stale.
05Restrict AI-written outreach sentences to four types only — fact (source-backed), inference (hedged with 'may/might'), offer claim (provably deliverable), or question (non-presumptive CTA). Any sentence that can't be classified into one of these should be deleted; this rule eliminates fake personalization.
06Email and LinkedIn approvals must be separate switches in the review interface — approving one channel does not unlock the other, and any message edit re-locks that channel until re-approved.
07Hard rule: any prospect reply must pause every active outreach channel before classification happens — never let a model decide whether a reply is important enough to halt other in-flight sequences. After pausing, classify (positive/objection/referral/unsubscribe/etc.) but do not auto-send substantive sales replies.
08Autonomy should be added one layer at a time and never two variables changed simultaneously: automate research first, then monitoring, then low-risk actions, keeping high-risk/external actions gated behind explicit approval until the layer beneath proves reliable. Judge the system by decision agreement rate, not message volume.
Adjacent fields
Latest evidence
J.B.
A detailed workflow for building an outbound sales agent using Hermes that connects Apollo, Clay, Crustdata, HubSpot, Telegram, Smartlead, and Unipile into a research-to-outreach pipeline: market->research->signal->message->approval->action->CRM, with staged autonomy, strict evidence rules, and hard stop conditions on replies.
Scott Schindler
Gumroad founder Sahil Lavingia runs much of the business through a Hermes agent called Gumclaw, built on the Fable 5 model. A public operating-system site documents how cron jobs, filesystem-based memory, growing skill libraries, and a tiered permission model let the stateless agent perform real support-to-engineering work autonomously.
EP
A workflow for turning a self-hosted Hermes agent into a lifelong knowledge system: instead of relying on capped built-in memory, memory acts as an index pointing to an Obsidian-style markdown vault on a VPS. Six copy-paste prompts build the vault structure, a self-filing skill with contradiction handling, an evidence-card answering method, an inbox-based braindump/import pipeline, and a weekly staleness-check cron job.
Nous Research
Nous Research's Hermes Agent page aggregates 262 real-world deployments scraped from X, GitHub, Reddit, HN, YouTube, and Discord. Stories span dev workflows, personal assistants, trading bots, enterprise deployments, and self-hosted setups — revealing Hermes as a general-purpose agent OS with self-improving skills, three-layer memory, and natural-language cron as its signature primitives.
Ole Lehmann
Ole Lehmann ranks 13 integrations for the Hermes personal agent, with Firecrawl (clean web data, fewer tokens), Browserbase (real browser sessions), Google Workspace, GitHub, and Obsidian as the essential five covering ~80% of needs. Composio handles one-click setup.
JUMPERZ
A two-layer agent workflow: Discord handles plain-English commands and task visibility (especially mobile), while Hermes Kanban is the execution system that assigns, tracks, runs, logs, and proves work. An intake bridge connects both, creating Discord commands → Hermes tasks → Discord task board mirroring.
Shann³
Comprehensive operator guide for Hermes Agent by Nous Research—an open-source autonomous agent framework with 150K GitHub stars and #1 OpenRouter usage. Covers architecture (brain/personality/skillset), four-level deployment progression (laptop to full VPS fleet), multi-agent control room patterns, a 21-step SEO pipeline, and the prototype→production methodology for marketing automation.
Garry Tan
Garry Tan describes a repeatable 4-agent build pattern: do a task manually, skillify it, schedule via cron, verify it's resolvable, add evals and integration tests, then repeat. The quoted framing treats agent failures as organ failures rather than model failures—memory, actions, and self-check are separate subsystems to diagnose and patch independently.
Alex Finn
Alex Finn's workflow uses one always-on Mac (Mac Studio) as the sole code-writing machine, with all other devices (iPad, iPhone, additional Macs) acting as remote command nodes via Codex's 'control this Mac' / 'control other devices' settings plus Tailscale for a private mesh network. The claimed result is location-independent coding from any device worldwide.
Matt Van Horn
A detailed operational guide for running AI agents across multiple Macs (daily driver + always-on Mac mini) using SSH aliases, Tailscale private networking, identical folder structures, and AgentCookie for browser session sync. Covers ops folders with CLAUDE.md, launchd for service persistence, and a full setup checklist.