AI + Web3 intelligence for builders and creators

Issue #008  ·  June 22, 2026  ·  shipguild.com

This week: Trav finally installed Hermes, and got a free, brutal masterclass from Quantic on what every beginner overcomplicates. Plus: Hermes v0.17 just shipped with iMessage support, the Fed shut the door on rate cuts, and Claude Code added /rewind.

From Friday's "AI Agents 101 with Hermes" stream — the most painfully honest beginner's guide we've done

Friday's stream was the conversation every "I'll set up an agent eventually" person needs to hear. Trav finally pulled the trigger on Hermes, and got a real-time, no-mercy masterclass from Quantic on every wrong turn he took along the way. By the end, the entire 90-minute journey collapsed into one prompt template that does what three custom-built agents and a week of stress couldn't.

Key takeaways from the stream:

An agent is just three things: LLM + tools + memory. Plus a decision-making loop on top. That's it. If you've used ChatGPT with web search and persistent memory, you've used an agent, most people just don't know what to call it. The terminology matters because it shapes which tutorials, docs, and YouTube videos actually help you.

The security panic is mostly misplaced. Q's reframe: it's not about the agent being dangerous, it's about general security hygiene. The right answer isn't "don't install Hermes." It's "don't keep your life savings on the same machine you use for daily browsing and AI agents." A $400 used laptop dedicated to crypto and bank logins is the move. Then run agents wherever you want on your main machine.

The biggest beginner trap: building Claude Console agents instead of using Hermes Desktop. Trav spent a week down that rabbit hole, environments, memory stores, MCP configuration. It's powerful but built for developers. Hermes Desktop (released June 2) is the actual beginner path: native UI, no terminal, plug in a model, type a prompt, done.

The pricing alpha most people miss. Trav was paying $20/month for Claude Pro + topping up Anthropic API credits to run Hermes, paying twice for the same compute. The Nous Portal plan is $20/month and bundles access to 300+ models from every frontier lab, plus web search, image generation, browser automation, voice. One subscription replaces both. Worth doing the math on your current stack.

The "Soul MD" trap. Trav had built three custom agents (Scout, Packager, Lens) each with hand-written context files, for what turned out to be a five-second prompt. Q's call: don't pre-engineer the system. Let the agent figure out the system. Premature optimization in agent land is the same disease as in software engineering, and it eats the same week of your life.

The "agentic laziness" prompt template Q demoed live:

The output is what Trav got back in seconds: a structured research report mapping news topics by viewer search volume (Claude/Anthropic at 45.5M monthly searches with 1,500% growth, OpenAI custom chip as a low-search B2B story, AI companions as a sleeping consumer giant) plus a recommended posting strategy. No custom agent. No MD files. No three-agent orchestration. One session, one prompt, one result.

"Don't try to figure out the system. Let the agent build the system. You just provide the goal, who you are, and what your day-to-day looks like. The thing will figure it out."

— Quantic, mid-rant, mid-stream

The Big Idea: Most people approaching agents for the first time treat them like complicated software they need to configure. The mental model that actually works is to treat them like a new hire — give them context (who you are, what you do, what you're trying to achieve), give them a goal, and let them figure out the system. The optimization happens through use, not pre-planning. Every hour you spend on custom MD files before you've validated whether the agent can solve your problem with a simple prompt is an hour you're not learning what the agent is actually good at. The configuration era is for after you've found the bottleneck — not before.

What to do this week:

1. Download Hermes Desktop for Windows or Mac.

2. Sign up for the Nous Portal Plus tier ($20/month → 300+ models + tools). Cancel any Claude API top-ups you're double-paying.

3. Open a session. Paste the prompt template above with your actual goal. Watch what happens.

4. If you don't have a dedicated machine for sensitive accounts yet — order a cheap refurbished laptop today. Move all crypto and bank logins to it. Use your main machine for everything else, including agents.

📺 Full stream: HERE  ·  🎵 TikTok clips: HERE

The non-terminal way to actually start using a real AI agent this week

Hermes Agent Desktop (v0.17.0)

Native Windows / Mac / Linux app from Nous Research — self-improving AI agent with persistent memory, tools, and now background subagents

Hermes Agent itself has been around for months, but always required terminal use and manual provider setup. The desktop app changed that — it shipped on June 2 as a native cross-platform front end, and v0.17.0 (June 19) added enough capability that it's the cleanest beginner entry point for anyone who's been postponing this exact step. The whole stack is free and open-source (MIT licensed). You pay for inference, not the agent.

The pricing question, settled:

If you're already paying for a Pro plan + topping up API credits to run Hermes, switch. If you're net-new, start here.

Getting started in 3 steps:

1. Download Hermes Desktop from hermes-agent.nousresearch.com.

2. Sign up for the Nous Portal Plus plan ($20/month, 300+ models).

3. Open a session and paste Q's prompt template (Section 1) with your actual goal. Don't build a system. Let the agent build the system.

⚠️ Honest caveat: Don't install on a machine that holds private keys, seed phrases, or banking credentials. The agent itself is fine, the security hygiene rule is "no agent on the machine that holds the irreplaceable stuff." A $300–500 refurbished laptop dedicated to crypto/bank logins solves this for life. Also: the desktop app is only three weeks old. Expect occasional rough edges; report them.

Download Hermes Desktop →

If Growth Bytes is part of your weekly stack, the best thing you can do is share it with one builder or creator who needs it.

SUBSCRIBE HERE!

🐦 X: x.com/shipguild (formerly @web3matters_xyz)
📺 YouTube: HERE
🎵 TikTok: HERE
🌐 theshipguild.com

We drop every Monday. See you next week.
— Trav & Q, Ship Guild (formerly Web3 Matters)