JBS Weekly

This week I covered five things you could try today: a pile of free Claude repos, an AI agent for your shop, an agent with a browser, Meta's new model, and Claude Skills. Different stories, same moment. You say yes, and something new starts acting with your access. This issue is about that moment, and a card that slows it down.
This Week’s Signal

Every tool this week arrived free or cheap, and the real cost landed after the yes.
The cleanest case is Anthropic's own testing. Claude Haiku 4.5 was running an automated test on random websites and filled in the form on a real unsolved-murders site with a made-up tip on July 18. The tip was marked as spam and never reached investigators. Police said they only learned of it when Anthropic told them.
The quieter versions: cloning a repo is an install, and a star count measures attention, not safety. Anthropic's enterprise guidance tells reviewers to read a whole skill folder, bundled scripts included, and run the scripts in a sandbox. The agent pitches were offers of access to your calls, your ad account and your customer files.
The model adds a bill. Muse Wonder's API list price is $4 per million input tokens and $20 per million output tokens, yet an early subscriber to Meta's paid plan reported rate-limit errors within minutes. The sticker price was never the cost. The quota was.
So trying something is a decision about access and limits first. Before you say yes, decide what it can touch and how you will know it worked.
The Playbook: The Trial Run Card
The daily posts gave you settings to lock down. This is the step before them: one card per new thing, filled in before you run it. If you cannot write a line, you are not ready to start.
Job. One sentence, one recurring task. "Draft follow-ups for unpaid invoices" counts. "Help with the business" does not.
Desk. Where it works during the trial: a throwaway folder, a copy of the data, a test account, a repo checked out away from your real files. Your live accounts wait until the review.
Pass mark. Write the number that decides it before you run it. Corrections needed per ten tasks, minutes saved, or cost per finished task. For a model, count finished tasks per day inside your plan's limit, not tokens.
Review date. A calendar date three to seven days out where you choose keep, cut or widen access. Access only grows at the review, never in the middle of a trial.
Here is the card for trying Muse Wonder on invoice follow-ups. Job: draft a follow-up email for each unpaid invoice. Desk: ten old invoices, nothing sends. Pass mark: no more than two drafts out of ten need a rewrite, and the batch finishes inside the plan's limit. Review date: next Friday. If it misses, it goes back on the shelf and cost you a week of watching, not a customer.
Use the same card for a repo, a skill, an agent or a model.
From The Podcast
All five episodes from the week, in case you missed one:
Claude GitHub Repos: Vet Them Before You Run Them: which free repos are worth a look, and why a clone is an install.
AI Agents for Your Small Business: 4 Guardrails Before You Say Yes: what owners actually buy, and what to demand first.
Before You Give an AI Agent a Browser: 4 Settings to Lock Down: the fake police tip and why a prompt is not a safety rule.
Meta Muse Wonder Costs the Same as Opus 5.5. Should You Switch?: where it earns its place and where the quota bites.
Claude Skills Explained: Teach Claude Your Business Once: build a skill from your own best and worst results.
Tool Worth Trying
The Trial Run Card starts with a job, and the AI Readiness Checklist helps you pick a good one. It is completely free. Answer ten yes-or-no questions about a process you repeat, and in about ten minutes you know whether it is ready to automate or needs cleanup first. Run it on one process, then fill in the card for the tool you want to try.
Joe’s Take
I learned the Trial Run Card the annoying way. I restarted my own server in the middle of a podcast render, and the episode errored out. Nothing was hacked or leaked. I just gave myself a button with no brake on it.
So I built the brake: the server now refuses to restart while the podcast pipeline is running, and nothing gets dubbed or paid for until I click approve. That is the whole card in practice. A job, a place it can run, a rule for when it's allowed to spend, and me deciding at the end.
The thing I keep relearning is that the risky moment is never the install. It's the afternoon three weeks later when it's been fine every day and you stop looking. Write the review date down.
Tools I Use
n8n — The automation tool I use to connect apps, trigger workflows, and stop doing things manually. If there's a repetitive process in your business, this is where you start fixing it.
VoiceInk — A local AI dictation tool for Mac that transcribes your voice with near-perfect accuracy and runs entirely on your device, meaning nothing you say ever touches a cloud server.
Blotato — Handles the full content distribution side of your business: drop in a topic and it generates platform-specific posts, or feed it existing content and it repurposes it across formats. TikTok videos become tweets, podcasts become blog posts. Includes a scheduling calendar, visual creation tools for carousels and infographics, and publishes natively to 9 platforms with no per-post fees.
Beehiiv — What you're reading right now is published on Beehiiv. If you're thinking about starting a newsletter or moving off a clunky platform, this is the one I'd recommend. 20% off your first 3 months with my link.
Google Workspace — Beyond email and Docs, a Business Standard plan includes Gemini Pro built into every app, NotebookLM Plus, and access to the enterprise versions of the whole suite. Better value than a standalone Gemini subscription when you're already paying for Google anyway. 14-day trial and 10% off your first year.
Descript — Video and podcast editing that works like a text document. You edit the transcript and the media follows. Cuts filler words, cleans up audio, and handles captions automatically. 50% off your first two months on the Creator Plan.
Final Thoughts
Five different stories this week had the same hinge: the moment between "this looks useful" and "it is running on my stuff." A week of evidence says that moment is where the risk and the cost both start. Slow it down with a job, a test area, a pass mark and a date, and most of the rest takes care of itself.
PS: Pick one tool you said yes to this month and write its card today. If you cannot name the pass mark, that is your answer.
Cheers,
Joe
