
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Hermes Agent is pitched as the new open-source agent to beat — the host says Open Claw “changed the trajectory of AI,” but argues Hermes Agent from New Research has now overtaken it with a local-first setup, stronger memory, and better day-to-day usability.
Local-first is the big practical advantage — Hermes runs on your laptop, uses your own API keys, stores everything locally, and lets you chat through terminal, Slack, or Telegram without the setup pain the host associates with Open Claw.
The dashboard is where Hermes really pulls ahead — built-in analytics, logs, cron job visibility, skill browsing, plugin installs, model controls, and API key management save “hours” versus assembling similar visibility yourself in Open Claw.
Multi-agent profiles make specialization feel native — the host shows seven profiles, including a designer on Opus 4.7, a cheaper Anthropic Haiku-based runner, and a custom ‘This Week in AI X writer’ with its own soul.md personality prompt.
Hermes ships with serious workflow surface area out of the box — it can code, run shell commands, manage files, search the web, control the browser, automate cron jobs and webhooks, and tap 80+ integrations like Spotify, Slack, GitHub, Notion, and Linear.
The Kanban board turns Hermes into a semi-autonomous task manager — a cron job checking every 10 minutes can triage tasks, rewrite vague prompts into fuller briefs, move cards like ‘research competitors’ to ready, and keep work flowing without manual board maintenance.
The video opens by giving Open Claw its flowers: when it launched, it “totally changed the trajectory of AI,” especially because one developer built an open-source autonomous agent people used for everything from outbound lead gen to content. But the whole point of this video is that Hermes Agent may have already made it feel old.
Hermes Agent is introduced as a local-first agent from New Research that runs on your laptop, uses your API keys, and stores everything locally. The host frames the quick start as dead simple: grab the install command from hermes-agent.newresearch.com, paste it into terminal, then run Hermes setup to connect keys and chat surfaces like Slack or Telegram.
What really won the host over wasn’t just setup — it was Hermes’s memory and its ability to auto-generate skills from conversations. He says his standard move with any new AI tool is asking, “What can you help me with? What are you best at? How could we work together?” and Hermes gives genuinely useful workflow ideas instead of vague fluff.
The capabilities list is long and concrete: coding, shell commands, file management, browser control, research, cron jobs, webhooks, PDF generation, and multi-agent orchestration. On top of that, Hermes launches with 80+ integrations including Spotify, Slack, GitHub, Notion, and Linear, which the host presents as the reason you can get to real workflows fast.
This is where the host’s enthusiasm spikes. He says Hermes’s dashboard is such a step up that unlike Open Claw’s native dashboard — which he says he “pretty much never looked at” — this one is actually useful, with sessions, connected platforms, analytics on token usage and models, logs, cron jobs, 87 built-in skills, plugin installs from GitHub, and even theme colors.
The most exciting feature for him is the profile system. He shows seven profiles, with Kimiko 2.6 as the default workhorse, a designer using Opus 4.7, a researcher, reviewer, runner on Anthropic Haiku, and even a custom “This Week in AI X writer” trained on his preferred Twitter/X style through a soul.md system prompt you can edit directly in the dashboard.
One of the most underrated wins, in his view, is simply seeing and managing all the secrets and API keys Hermes is using. He contrasts that with Open Claw, where wrangling keys was “a nightmare” because it was hard to tell which service needed what or which credential had broken.
The closing demo is the Kanban board, one of his favorite parts of Hermes. Tasks can begin as rough ideas in triage, then move through to-do, ready, in progress, blocked, and done — but the clever part is a cron job that checks the board every 10 minutes, enriches thin tasks like “research competitors,” moves them to ready, and writes a more complete brief with context, deliverables, and suggested analysis steps.
Share
Keep Reading
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.

Playbook
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.