Brian Casel

This will be a good one! Going live in about ~30 mins

Spending a lot of time iterating on how bots send me stuff. They're only as productive as I can manage them!

Before: Almost dinner? Let's wrap this up and ship. Now: Almost dinner? Let's plan 3 (big) things to queue up for the orbs night shift...

Giving each of my Grok Bots an email address (agentmail) feels like a huge unlock. The chat experience is nice, but chaotic and hard to be productive with a "team" of bots. Emails give me exactly what I need to receive reports, approve decisions, get progress updates, etc.

If you're curious about building with orbs in the cloud, I'm running a live workshop, demo & Q&A today at 2pm ET. Join us!

The idea of a "super app" for work with AI doesn't work for me. Different tools for different jobs: Building with cloud agents: Amp Hand-writing & editing: Flywrite Deep research projects: Codex Background tasks: Grok Bot Strategic thinking: ChatGPT & Claude

This is how we work now. Multiple projects happening. Each orchestrated by a lead agent delegating to sub agents. Each working on their own machine (orb) with @AmpCode. What a time.

I miss freely using an mdash when it's actually warranted.

So much of my multi-tasking is just preliminary research: "Here's a complicated thing that needs to happen soon. Research the complexities, consider our options, and map out what our starting point should be." Then I cycle back around and dive when I'm ready to "get into it".

Launched a refresh of buildermethods.com today (on it's 1 year birthday 🥳) It's a return to it's original promise: Helping fellow builders stay ahead as our craft is changing faster than ever.

All my work happens on orbs (cloud agents) now. I work working through one "orchestrator" who spawns a few team members (code implementer, copywriter, designer, etc.). Each on their own orb. Sometimes in different repos. Often using different models. All talking and coordinating with eachother.

Building with cloud agents is... a big deal. My next live workshop is Oct 1st where I'll show how I've moved ALL my work to the cloud. Projects, tools, multitasking, agent orchestration... All of it. Grab a seat here (free for BMP members): buildermethods.com/sessions/bui...

Shipped a 'manage my subscriptions' UI for mimeohq.com

I'm already doing all my work from the cloud (orbin' on @ampcode.bsky.social). Working from mobile more than ever now. The idea of the iPhone Duo is getting more appealing 🤔

Just about every product I want to use (and build) needs: - App for UI, database, and deliverables. - MCP for agents to read and write to it. - Repo for working artifacts, agent training, & configs. - Agent skills for getting users' input.

I'm not a first-gen guy. But I'm surprised how appealing the Duo seems to me. I like how small it seems folded. Like the idea of more space when working from mobile. Moving the dock to the sidebar feels right. Still. Not a 1st-gen guy.

So... iPhone Duo or Pro?

Who's creating video content? I'm exploring a new idea and I'd love to chat. DMs open

This year I’m removing templates, scaffolds, and libraries from my process. I'm realizing they're no longer shortcuts. They're constraints. AI models are improving so quickly that forcing them into old patterns means missing out on much of what newer models can help us do.

fyi - X is a lot better now. You should go back.

How I trust code I never read: 1. Heavy up-front planning sessions, PRD, specs 2. Agent self-verification loops before delivery 3. Manual QA on critical paths only No surprises. No hallucinations. A few refinements, then ready.

New in flywrite.app: I added Cursor CLI integration so it knows your current file and selection (just like it does Claude Code and Codex CLI). Now I can use my favorite markdown editor (Flywrite) with my Cursor/Grok subscription.

All videos for my next course, Become AI-Native, are recorded and in the can. Editing, uploading now... Launching later this week in Builder Methods Pro. (pssst: Raising price for new members after launch, now's a good time to grandfather in)

Released another free tool: Skill Builder. A skill for creating skills! Includes my favorite conventions and strategy for turning processes into skills for your agents. buildermethods.com/skill-builder

Here is agentcanon. Simple, lightweight symlink setup so that claude and others read from .agents/skills and agents .md. Also comes with a handy HTML manifest to catalog all your skills, their locations, spot duplicates etc. Install it with a prompt: github.com/buildermetho...

Using skills and claude.md/agents.md continues to be the most annoying "lock-in" thing. Claude more to blame here for not embracing 'agents' standard naming. Plenty of hacks and workaround strategies, of course. All annoying (and even harder to teach newcomers). 🙄

Working on my next course, Become AI-Native (for those new to AI-assisted workflows). Hardest part to teach: Harnesses and that rapidly changing landscape. Trying to make it practical & future-proof. ChatGPT (most popular brand name) not making it easier with their inept naming and product UX.

Writing and building my next course, 'Become AI-Native'. flywrite.app is the perfect tool for this type of work (writing-heavy project with some design work). Developing & iterating with Claude, hand-editing some parts...

Last year I created Agent OS to keep coding agents on the rails. This year, I stopped using it. Today's models do worse when you micromanage how they write code. They do better when you pour everything into planning and verification.

As much as I build and ship lots of stuff (public products and many internal tools)... The big trend for me has actually been stripping back and simplifying a lot. Agents and speed of iteration means a lot of heavy-handed tooling & process can fall away.

I've been extremely anti-meeting my whole career. My teams and I always worked 99% async. Now most of my day is in meetings planning big projects. I'm the only human at these meetings. Work output? Unprecedented on all metrics. No question. Enjoyment and flow? Open question.

How to make a Slack killer: Let users log in and find their workspace without requiring 8 attempts with different emails and workspace names.

The thing that disqualified non-technical founders (can't code) has been neutralized. What you have (deep knowledge of a real problem) can't be shortcut or replaced by AI. You're not behind. You're way ahead.

I built it. This is mimeohq.com Email marketing you own. Your list. Your flows. Your emails. Just plug in a sender (Resend, Postmark, or SES). Manage it all in the UI or with agents. Opening soon. See demo video 👇

The difference between vibe coding and building: Vibe coding is prompting and hoping. Building is a methodical, predictable workflow. No surprises. Just your vision, made real. See my entire building process here. Free. https://buildermethods.com/workshop

There are obviously a ton of amazing tools that have helped us build Eventually, but I want to call out one especially: @briancasel.com Builder Methods. The community, the videos, the workshops, but especially the tools that Brian shares have all been extremely crucial for us.

Product built. Now writing copy to explain it. With flywrite.app of course.

Spruced up my Claude Code status bar today.

Sales slowed down for a little while there. Then I remembered I hadn't updated my email sequences and automations in a while. Turns out customers still read emails.

First time in a while I'm optimistic about starting software products again. Seeing a new shape emerge: Familiar problems and tool categories solved in a way that gets accelerated by AI adoption instead of eaten by it. Requires rethinking product shape and biz model. Marketing still hard though.

Domain purchased. Gonna build the thing I've talked myself out of building for 15+ years.

If you feel like you have to review every line of code AI writes, the bottleneck isn't AI. It's micromanagement. Move that energy into planning and verification. Own the what and the proof. Let the agent own the how. That's what unlocks bigger builds.