From the Weekly Meeting

Is AI Safe for My Family?

AI is safe for most families with a few ground rules: use parental controls, keep sensitive chats supervised, and teach kids to double-check answers.

Are clients done with the one-size-fits-all SaaS model?

• Collin Thomas

Collin's read on the market: everyone is used to the SaaS model, one way to do it, take it or leave it. Customized is the hot thing now. Give a client exactly what they need and that's what's special; white-labeling locks you back into defining one way.

Can a bedtime brain dump count as a real project spec?

• Gina Wang

Full disclosure: she did it while lying in bed, falling asleep. This is what I want, I'm going to sleep, don't ask me any questions. The bedtime brain dump is now a valid spec.

Can one dropped prompt replace twenty onboarding questions?

• Frank Priboy

Frank's path to a sellable second-brain product: take the extraction prompt, flesh it out, and wire it into the interview process so the client does the work and the system maps it. Matthew's version pulls a client's tech specs through one dropped prompt, no 20 questions, nobody feels dumb.

Can people tell when they're being served instead of extracted from?

• Collin Thomas

A psychology layer for onboarding interviews built on nonviolent communication. Probe gently: people can feel the difference between being served and being extracted from, and they respond accordingly.

Is there one kind of product that can be white-labeled after all?

• Collin Thomas

The exception to never white-labeling: a product that interviews the user and molds itself to them could be white-labeled, because the customization ships inside the product. If you can build that, it sells everywhere.

Is your Claude history the best source material you already have?

• Matthew Sutherland

Your Claude history is a first-person record of what you know, how you talk, and what you care about. Gina's engine reads her transcripts as its primary idea source, and Matthew calls it the most valuable piece of the whole design. Your voice is already on file. Use it.

Should you keep the maintenance contract or hand over a runbook?

• Frank Priboy

The build decision at handoff: keep maintenance as a revenue stream, or one-and-done with a runbook. Frank's requirements for the interview-driven build: private enough that people talk honestly, full enough that the system learns what they really need, not just what they think they need.

What can you build in fifteen minutes of talking before you fall asleep?

• Gina Wang

Gina talked to Fable for fifteen minutes while falling asleep and woke up to a rebuilt website, eight posts in her voice, and a working content pipeline. Regular effort, no Max mode. The models need a clear goal and good source material more than they need a perfect prompt.

What changes when your onboarding interview actually listens?

• Matthew Sutherland

Onboarding interviews with dynamic follow-up questions and Fable-level reasoning behind them. The script probes based on what the person actually said, not a fixed question list.

What does a working content engine look like from the inside?

• Gina Wang

Screen-share walkthrough of Gina's content engine: her own transcripts as the idea source, rubric gates on quality, and one human approval point in Slack. She named the AI staff Birdie and Lanny. One of them rejects her ideas.

What happens when you ask an autonomous workflow when to publish?

• Matthew Sutherland

Matthew asked when he could publish the LinkedIn post his Fable workflow drafted. The answer: it's already published. Audit, red-team, image, approval, publish, all in one autonomous run.

What happens when you write Dale Carnegie into your prompt?

• Matthew Sutherland

Dale Carnegie, written into the prompt. Matthew bakes Carnegie principles into the interview script so the questioning is sincere and authentic by design. Applies to sales calls and AI prompts alike.

What keeps a content engine running after the first build?

• Gina Wang

The flywheel behind Gina's engine: study what performs, draft in her voice, gate with rubrics, publish, measure, repeat. Even 1% a day beats nothing when you're measuring from zero.

What makes a system feel helpful instead of extractive?

• Matthew Sutherland

You are authentically helpful and not extractive. The closing thesis of the call: build interviews and systems that serve people instead of extracting from them, because people can tell.

Why does a perfect technical SEO score still rank for nothing?

• Frank Priboy

Frank's SEO audits score hundreds across the board on the technical side, and he knows that's not enough. Perfect structure with no content ranks for nothing; great content without the tech never gets seen. His next move: blend his technical audits with a content engine.

Why does a white label mean you maintain it forever?

• Gina Wang

A white label means you maintain it forever. A productized system with a playbook means they own it and you move on. Gina asks how to productize during every build, so the playbook writes itself alongside the work.

Why give your AI setup away instead of white-labeling it?

• Matthew Sutherland

Matthew's take on white-labeling an AI operating system: everything moves too fast and every setup is too personal. His Athena would not work on someone else's stack. So he gives away test setups to newer builders instead and learns how beginners actually onboard.

Are your shared Claude links showing up in Google search results?

• Matthew Sutherland

Some publicly shared Claude apps, documents, spreadsheets, and other artifacts are appearing in Google search results. Although Claude content is private by default, anyone who creates a public sharing link makes that content publicly accessible. If the link is then posted somewhere visible to search engines, it may be indexed or archived.

How do you move work from Claude Code to Cowork without losing context?

• Gina Wang

The pattern for moving work between Claude Code and Cowork: ask the current session to summarize what exists, then have it write the prompt that briefs the next tool. Let the AI explain itself to itself. Gina shows the handoff in under a minute.

How do you run AI email triage when Proton blocks the bridge?

• Matthew Sutherland

Proton's security layer blocks AI tools from driving its desktop bridge, which kills automated email triage. Matthew's workaround: open Proton Mail in the browser and let Claude for Chrome work it directly. Same triage, no fight with the bridge.

What does it mean when your AI task stops to ask a question?

• Michael Wacht

Michael asked Cowork to tailor his resume for a job posting as a side thought while working on something else. Instead of a done-marker, the task raised a little hand: it had a question about where the latest resume lived. He pointed it at his Obsidian vault and got a near-perfect job-specific resume back.

What does the finished state of email triage look like?

• Frank Priboy

Frank's target state for email: seven or eight scattered inboxes all forwarding into one labeled Proton box, with analysis happening at the individual email layer. Active clients, family, and priority senders break through to Slack. Everything else waits until he checks.

What happens when you bet on the channel you're building?

• Frank Priboy

Frank's cousin Joe Santagato turned down a butcher-shop job from their uncle fifteen years ago because YouTube was already paying him $60K a month. The Basement Yard went on to sell out Madison Square Garden. Bet on the channel you're building.

What if publishing a Claude artifact is free distribution, not a leak?

• Frank Priboy

Publishing a Claude artifact puts it in the search index. Frank flips that from a privacy problem into a distribution channel: anything you actually want found gets free organic reach the moment you hit publish. The rule is knowing which side of that line your content sits on.

What's missing from a pipeline that hands you two hours of reading?

• Jacob Brodsky

Jacob's overnight pipeline researches while he sleeps and hands him two hours of reading every morning, which is a pipeline with a missing final stage. The fix on the table: migrate it into Cowork and have Claude Code write its own handoff prompt. Output should end in a decision, not a reading list.

Which repo security scan should you run first?

• Frank Priboy

Anthropic's repo security tool runs in two steps: a manual scan in each repo first, then the background plugin that keeps watch. Frank walks the setup and why /doctor comes first. Codex shipped a security CLI the same morning and GitHub is rolling out its own scanner, so layer them.

Why define the mission before you build the system?

• Michael Wacht

Define the mission before the system. Michael's is daily efficiency: digest the day in ten minutes, take one thing away, preview the next five days. A clear mission statement is what tells you when a pipeline has a missing stage.

Why do your completed tasks keep coming back from the dead?

• Michael Wacht

Michael's completed tasks kept resurrecting because Obsidian stores files, and files have no state. He wired Todoist's free tier into Cowork through the API and the zombie task list died in an afternoon. His line for it: quantum entanglement for to-do lists.

Why rebuild a to-do list the market already perfected?

• Michael Wacht

Why build a to-do list when one already exists that just needs seamless integration? Michael names the builder's trap: making tools the market already perfected while avoiding the real work. His honest version: everybody's hiding out from cold calls and sales. That's why we build.

Most people start building with AI too fast. Here's the habit that fixes that.

• Mike Thomson

Mike Thomson shares a practical method that consistently produces better AI-built projects: write a project brief first, then use "planning mode" before you touch a single tool.

Prompt injection explained -- what it is, why it matters, and why your AI keeps flagging things it shouldn't.

• Collin Thomas

Colin Thomas breaks down one of the most misunderstood concepts in AI security in plain language. No jargon, no fluff -- just a clear explanation of what prompt injection actually is, how attackers use it to manipulate AI systems, and why the same safety filters designed to stop it sometimes block perfectly legitimate technical and security-related language.

She built a voice AI agent that coordinates group travel -- and finished 3rd at an invite-only hackathon with 400 developers.

• Gina Wang

Gina Wang shares the full story from an exclusive travel industry AI hackathon -- what she built, how she built it, and what the experience revealed about where voice AI and agentic systems are actually headed.

What happens when AI doesn't just assist it executes?

• Ron Nedd

Ron Nedd walks through a live demo of Cowork's "Record a Skill" feature, showing exactly how he turned his own real estate workflow into a reusable AI skill no coding required.

What if your AI remembered everything -- and briefed you every time you sat down to work?

• Jacob Brodsky
• Josh
• Mike Thomson

Jacob and Mike Thomson walk through how they each set up persistent memory AI workflows, also called AI second brains, and what that actually looks like in daily use.

Are you using system-building as an excuse to avoid real work?

• Michael Wacht
• Collin Thomas

Michael Wacht and Collin Thomas discuss the trap of endlessly tweaking and building your organization system instead of actually doing the work.

Can you safely use AI on intellectual property you can't afford to leak?

• Herman Moore
• Collin Thomas
• Michael Wacht
• Matthew Sutherland

Yes, but not with public models like ChatGPT or Google -- protecting IP you can't afford to leak means keeping the sensitive parts on infrastructure you control. Guest Herman Moore explains why he won't put his patented IP into public AI tools, and Collin Thomas, Michael Wacht, and Matthew Sutherland break down the actual architecture needed for data privacy: local models, hardware requirements, and deciding what runs on-machine versus what gets sent out.

How can AI help people who aren't developers?

• Dena
• Michael Wacht

AI can help people who aren't developers by working as a practical, jargon-free tool for everyday problems, not just programming. When Dena, a writer dealing with writer's block, asks how AI could help her process, Michael Wacht gives her exactly that: a plain-language answer built around real life instead of code.

If you understand RAG, are you already too technical?

• Mike Thomson
• Collin Thomas

Mike Thomson and Collin Thomas discuss why understanding technical concepts like Retrieval-Augmented Generation (RAG) might mean you've lost touch with how normal people use tools.

Join the AI for Life community.

• Matthew Sutherland

Matthew Sutherland promotes the regular Wednesday session of AI for Life.

The one settings gotcha that will break all your file links.

• Michael Wacht

Michael Wacht shares a specific settings gotcha that will break your file links if you miss it -- the kind of practical tip that saves hours of frustration.

What does 'intellectual sovereignty' actually mean?

• Michael Wacht

Michael Wacht defines the concept of intellectual sovereignty -- maintaining control and ownership over your ideas and data in an AI-driven world.

What does a working AI operating system actually do all day?

• Michael Wacht

Michael Wacht provides a screen-share walkthrough of a functional AI setup. He demonstrates how files are structured and shows the concrete daily automation -- including email and calendar triage, week-ahead previews, pulling drive times into calendar entries, and an article-writing pipeline.

What does it actually mean to build a second brain?

• Mike Thomson
• Matthew Sutherland
• Collin Thomas

Building a second brain means moving past simple filing into a system where you actively capture, organize, and can retrieve what you know, instead of just storing it and hoping you remember where it is. Mike Thomson admits he's been too embarrassed to ask what everyone means by a 'second brain,' so Matthew Sutherland and Collin Thomas give him the complete beginner's breakdown, from basic filing systems to how a second brain actually functions in practice.

What happens when one formatting rule sends an AI into an endless loop?

• Matthew Sutherland
• Jacob Brodsky

Matthew Sutherland and Jacob Brodsky share a debugging horror story where a single formatting rule sent an AI into a six-hour endless loop at 2am, ending with the classic paperclip analogy.

What happens when you ask the wrong question about an AI tool?

• Collin Thomas
• Karen Widas
• Michael Wacht

Karen Widas asks a question and receives a product demo instead of a direct answer. Michael Wacht and Collin Thomas step in to re-diagnose what she actually needed to know, turning a confusing moment into a clear, practical lesson.

What happens when you roast the expert?

• Collin Thomas

Collin Thomas opens with a roast of Michael Wacht, turning a moment of conflict into a real, practical lesson about AI tools.

What is a second brain, and why is everyone afraid to ask?

• Mike Thomson
• Matthew Sutherland

A second brain is just a system, usually digital, for saving and organizing what you know so you don't have to hold it all in your head -- notes, files, and ideas kept somewhere you can actually find and use again. Mike Thomson admits he was too afraid to ask what a 'second brain' actually is, and Matthew Sutherland gives him that straightforward, one-line answer.

Why AI isn't just for programmers anymore.

• Michael Wacht

Michael Wacht makes the case that AI is no longer just for developers -- it's a tool for real life and everyday work.

Why backups are like flossing -- everyone knows they should, but nobody does.

• Collin Thomas
• Mike Thomson

Collin Thomas and Mike Thomson discuss backups, why nobody actually does them, and why they are absolutely critical.

Why building a system so you understand it means you're doing it wrong.

• Michael Wacht

Michael Wacht explains a contrarian approach to organization: if you build your system primarily so that you can understand it, you're building it incorrectly for an AI.

Why humans aren't going to read your data -- AI is.

• Michael Wacht

Michael Wacht delivers a tight explanation on why the future of data organization is about machine readability, not human readability.

Why running local AI is like being a PC gamer.

• Collin Thomas

Collin Thomas uses a vivid analogy, comparing the expertise needed to run local AI models to PC gamers overclocking their rigs.

Why you need to turn your training data off right now.

• Michael Wacht

Michael Wacht gives a 17-second warning on why you should turn off training data sharing in your AI tools, and what companies retain if you don't.

Why you should build your organization system for the AI, not for yourself.

• Jacob Brodsky
• Collin Thomas
• Michael Wacht
• Matthew Sutherland

Jacob Brodsky, Collin Thomas, Michael Wacht, and Matthew Sutherland discuss why your current organization system will inevitably need to change. They explain why you should keep your core method stable while swapping the underlying parts, and why building structure for a machine to read is more important than building it for human readability.

Why you shouldn't just dump your existing files into an AI.

• Michael Wacht

Michael Wacht explains why dumping your existing, disorganized files into an AI system will just create a mess, and shows what you should do instead.

Why your organization system is guaranteed to break.

• Collin Thomas

Collin Thomas explains why your current system will inevitably need rebuilding, and how to prepare for it when it happens.