A Brief Intro…
Welcome Back! It's been a heavy few weeks for releases: new models, new commands, and a lot of noise. So, this week I cut it down to the 6 tools I'm using right now, each with a guide attached. Most are pulled straight from my socials, so follow along there to stay current.
Email me anytime if you want to work together or have a question ([email protected]). Let's get into it.
Recent updates…
The big one, and it moved fast: Anthropic launched Claude Fable 5 on June 9. Its first public "Mythos-class" model, a tier above Opus 4.8, and disabled it for all users three days later under a US government directive. If you meant to test it this week, that windows closed.
Quick version: the US issued an export control order after another company claimed it could jailbreak Mythos, citing national security. Because the order covered all foreign nationals, including Anthropic's own employees, the company had to pull both Fable 5 and Mythos 5 for everyone to stay compliant. Other Claude models are unaffected. Anthropic is complying but disagrees, calling it a narrow jailbreak finding and saying it's working to restore access.
This is exactly why local model’s matter. A frontier model you depend on can vanish overnight by government order, vendor decision, or pricing change, and your workflows go with it. Nothing about your setup was broken, yet the rug's gone. An open-weight model running on your own hardware can't be recalled out from under you. It won't match Fable at the top end, but it can't be switched off either. The lesson isn't "go fully local” it's don't build your core workflow on a single hosted model you can't replace. Keep a capable local fallback so a directive in DC doesn't freeze your pipeline.
The command I'm leaning on most right now is /goal. You give it the end result you want, and it works in a loop — building steps, verifying each one itself, and pinging you when the goal is actually done. It changes the job from babysitting prompts to defining outcomes. A few things to know before you turn it loose (especially around setting check-in guardrails so it doesn't run too far on its own), so I put them in this guide: [link]
The good stuff…
The 6 tools I'm actually using right now, with their full guides:
Codex — OpenAI's agent that went from coding assistant to full desktop operator: it can control apps, browse, generate images, and run long jobs in the background. The closest thing to handing work to a junior teammate who actually does the steps. [guide]
OpenClaw — Turns a Mac into a 24/7 AI employee you delegate to over Telegram, not a tool you babysit. It runs a 5-agent stack and works your portfolio while you sleep — the setup is the hard part, so the guide matters here. [guide]
Higgsfield — One subscription, 15+ of the best AI video and image models (Sora 2, Veo 3.1, Kling) plus cinematic camera controls and viral presets. Draft on a cheap model, render the final on a premium one. [guide]
Obsidian — The free second brain where I store every prompt, workflow, and decision. It saves tokens and gives your AI tools the context to actually serve you well — better organization, better output. [guide]
Claude — The real power isn't the chat box, it's the ecosystem: MCP connectors plug it into your tools, Skills teach it to do real jobs (actual .pptx files, not text), and Claude Code runs the whole thing. [guide]
Gumloop — Drag-and-drop AI workflows that reason and decide, not just shuffle data between apps. Design from the output backward, test on one record, then let it run on a schedule forever. [guide]
Wrapping up…
That's the week. As always, if you've got questions about anything in here, AI or otherwise, reach out via my socials or email me directly ([email protected]).
One more thing: I just launched a YouTube channel where I'm documenting the actual building, the businesses, the tools, the life around it. If any of these have been useful to you, the best way to support it is to watch the first intro video and subscribe (Link). That's it. See you next week.
— Thomas