How do you like your markdown docs to feel? Dial in the perfect fonts, weights, colors, scale in flywrite.app. See mine here. Now available.

Early group sold out. Flywrite is now open 🚀 Launch special on during July.

Funnel analytics SaaS have always required heavy (expensive) custom dev work to track the right data, cohorts, and filters. Without that work, the reports aren't actionable and often misleading. Built my own with Fable today. During lunch.

First customers in the door! 🚀

First invites for flywrite.app are going out today! Beautiful markdown writing apps lack a terminal for agents. Coding apps are terrible for writing (if they even have file editing at all). Flywrite gives you both. Beautiful markdown writing + terminal for agents.

Favorite alternative to Wispr Flow for voice dictation? Too many papercuts lately. Looking for better support for dictionary, better context (know when I speak mid-sentence etc.) and better keyboard customizeability.

Stop hunting for app ideas. Start from friction. The best first tool (or agent skill) to build isn't something you invent. It's a problem you already have, done manually, repeatedly (and wish you didn't).

Line in the sand: When what you're working on is code, use your favorite agentic dev tool. When the file you're writing *is the thing* (docs, skills, strategy, a plan, a journal, a blog post, etc.)... Use Flywrite.

"You don't have to do this anymore unless you want to." -- Louie C.K. to himself. From his conversation with Rick Rubin on creativity, craft and career in comedy. Through all the work and rollercoasters and sacrifice, it's a claim of victory. Freedom. www.tetragrammaton.com/podcast/loui...

Build your way out of a task when: - it recurs - you're pasting between apps that should just talk to each other - you're paying for SaaS that does 10% of what you need Pick one with clear ROI. Build that first (this week).

I plan builds twice. For a reason: PRD = the macro plan. My product vision. The decisions around *what* to build. (👈 vast majority of my time is here) Plan mode = the micro plan. Claude's plan for how it will code that vision. (👈 this alone is not enough)

Good morning to you too, Fable.

Super cool idea to be able to chat with the agent while looking at the document. Great way to keep the agent on the rails and focused while writing.

Surprise: I built a Mac app over the weekend (while on vacation 😎). It’s called Flywrite. It's a markdown writing app with a terminal built in, for people who write with Claude Code (or other agents). See the demo and join early access here: flywrite.app

A frontier model can build your whole app in one prompt. Sometimes that's even true. The question is whether you should. And no, you shouldn't. One-shotting skips every checkpoint where you'd catch your own planning misses. Build in milestones.

"I'd build that, but I can't code". FYI: If you have insider info on a hard problem, you have a huge advantage. You have the piece that AI can't do for you (or for any "pro" software dev for that matter). Building is learnable now.

"It works", Claude said. Reality: All it did was re-read its code and decide it was "correct". Make Claude prove it: 1. Write tests, require 100% pass. 2. Use agent-browser skill to screenshot & check its UI work. That eliminates a ton of time having Claude fix stuff.

This answer to my surveys will never get old. So helpful.

"Don't rebuild unless you absolutely have to" ...was the conventional wisdom for years. We shipped compromises and papercuts because rebuilding didn't make business sense. Discarding code is cheap now. Your v1 is a draft, not a monument. Rebuild liberally.

My internal app for planning my pipeline, scheduling and publishing content. I develop pieces in Claude Code. Agent pushes to here then schedules things out, according to my rules.

I'm giving away my entire process and tools I use to build apps with AI. Free. Non-technical? This is how you go from vibe coder to builder: - 60-min workshop (idea → PRD → build → refine) - my PRD Skill - my Design System - my app template Available here: https://buildermethods.com/workshop

I find myself doing more "lazy" prompting lately. It's like a "fast-forward" button through a project. "Can you tweak my speaker notes for slides 6-20 based on the types of things I've had you tweak in slides 1-5?"

I hand-write 0% of my code. I miss it. The bright side: The craft moved up a level. From writing code to designing systems that write it. Same technical-creative muscle. Way bigger lever.

The tool (or agent skill) your business actually needs is already hiding in your week. You don't have to dream up a clever app idea. Look at the recurring busywork that quietly eats your hours every week. That's the thing worth building. Start there.

A stronger AI model isn't a threat to your job. The best output from the best models comes when someone with judgment, taste and experience is steering it. More horsepower goes to whoever's willing to migrate their technical chops UP the ladder to where they matter more now.

Moved all of my content drafting work into Claude Code recently. Claude's turnaround time when editing content in markdown is so noticeably faster than (even simple) coding tasks. A lot faster than Claude .ai too.

I want a voice dictation tool that: - Doesn't add a period when I pause to think about the next word. - Knows when I started speaking mid-sentence and doesn't capitalize the first word.

Photoshop: For all those tiny annoying design tasks that are impossible elsewhere but maddeningly over-complicated and make you feel like an idiot (yet you're still using it after 20+ years).

Nice benefit of moving back to Claude Code as my daily driver: remote-control. Leaving the office while it's still cooking. Checking status from mobile. (for better or worse).

My very scientific test of new AI models: How often I find myself saying "a few fixes please..." vs "Perfect. No notes."

Creative meeting with my video editor. 👀

"AI-native" isn't about automating yourself (or your team) out of your business. It's moving your time upstream—where judgment, taste, and strategic thinking matter more—while agents run the recurring production work. You don't disappear. You relocate to where people matter most.

A real product gap in Claude Code CLI is a proper queuing or steering system. Ability to write up multiple next tasks in a queue, edit and reorder them before they come up, or push a steering prompt into the current active task. Conductor and Cursor handle this really well.

Tip: Ask your agents for feedback. "What was confusing or added friction in the instructions I gave you?" "How can we improve the Skill you just ran so that you produce more reliable output next time?"

The ideas that do damage aren't your bad ones. You kill those on sight. It's the 6/10. Seemingly good enough to keep, so you grind it hoping to turn it into a 9 or 10. A real 9 or 10 writes itself. Less "rescuing" weak ideas. More green-lighting the high-percentage shots.

Stop asking "Can AI build this?" Ask "Can I specify this?" AI knocked down the syntax barrier. The hard part is still on you: defining the problem, the constraints, the milestones clearly enough that an agent can actually build it. That's the skill to develop now. Not prompting. Specifying.

Dear Internet: Stop using Medium to publish interesting articles I want to read. Or if you do, it better be so good it's worth me verifying with a throwaway email, going through 3-step onboarding and upsells before I'm allowed to read past paragraph 1.

Don't love "workflow" being a keyword for Claude now, which instructs Claude to use its' new Workflow tool. The apps I build are for managing MY workflow, so I often frame my specs/prompts around the user's "workflow" with the feature/UI. That's a different meaning/context.

Two (very different) ways I work with AI: 1. Building: Me working directly with Claude Code etc. using a spec-driven process. All day. Every day. 2. Delegating: Hermes agent running Skills (processes) on a recurring schedules. Talks to my apps and files, sends me updates.

The AI prompted playlist feature on Spotify is pretty great. Here are 2 of mine (in my work account):

Noticing Opus 4.8's writing style when summarizing the work it did (coding) is much easier to read. Axios-style, very scannable, just the facts. I like.

Confirmed — Opus 4.8 still loves the emdash.

Seeing how far I can go to make my solo creator business AI-native. Here's my slate of custom apps and agent skills that run my pipeline. Goal: Less of my time on busy work. More of my time on creating, teaching, building.

Setting up Hermes as my agent platform for running all background work. Really impressed. Also really liking Discord as the messaging tool. Better markdown, readability, threading support than Telegram and others.

Been working on some internal tools for me & my agents to power my content pipeline. So much freedom and power when designing for a customer base of one.

Feature request for all agent harnesses: Emoji reactions on paragraphs written by agents so I can approve/reject/confirm each thing. So much of the work now is sifting through long planning discussions. AskUserQuestion and numbering responses only take you so far.

My building workflow these days: 1. Long planning sessions in Claude. 2. Build a PRD from that (now in HTML!) 3. Break out to buildable milestones. 4. Agent builds + writes up notes for next milestones' agents to read. 5. Review & refine after each milestone. 6. Repeat.

Not everything needs to be super fast when leveraging AI in your work. I spend entire days and weeks shaping and reworking specs, design plans, drafts, directions (in collaboration with AI) before I let the agent cook.

Planning to redesign my video recording space. Walls, lighting, arrangements, set design... Who's got a nice setup? Looking for ideas.

My agents tend to run on this 3-part pattern I call 'The Night Shift': 1. Agent runs a Skill on a recurring schedule. 2. Reads/writes to a markdown file or custom app. 3. Sends me a report, I approve or give feedback. Examples in today's video: www.youtube.com/watch?v=eqbf...