Andrew Warner

Apple's dictation sucks, and they're kneecapping alternatives.

@heydannymiranda @ComedicBizman Hell, yeah

Today is a special day. I am launching a new podcast with my friend @ComedicBizman. I have no idea where this is going. But I know that every time I talk to my friend Tej, it leads to new insights, laughs, and a general feeling that all is well. It's audio only. Episodes will be published at 5am on Wednesdays for the next year. I hope you enjoy listening as much as we enjoy recording. Sending big love your way!

@lennysan @wadefoster has been demoing his cursor setup and makes me want to switch to it

Narrative violation: Finding it's more fun to work within Cursor than in the native Codex or Claude Code apps. Not a massive difference, but just enough to keep me there. And obviously easier to play with new competing models as they come out. Nice work, Cursor.

@elvissun Want

i built the world's most personalized podcast. every episode recorded just for me. every morning. 8:30am. it turns 100+ industry newsletters into a 15-minute audio brief on my phone. it's replaced ~2 hours of daily reading. and it took me 4 years to build this. not because the pipeline is hard. but because it's plugged into my personal knowledge base that took 4 years to curate: 3k notes. memory, daily logs, customer convos, past research, every piece of content I've written. so this podcast doesn't list 10 generic takeaways. it tells me exactly the 2 ideas that contradict, extend, or compete with my own thinking. it's like having a clone of yourself gathering information for you 24/7. and that's what makes each episode so interesting to listen to. the pipeline: stage 1 - ingest, 8:00am pure shell, no LLM. pulls every newsletter from gmail, strips zero-width and bidi chars, NFKC normalizes, screens for prompt injection, wraps the body in <> markers. trust boundary. nothing reaches a model yet. (this is the reason I punted on this idea for so long, the prompt injection risk is real. but the first 2 stages solve that with isolation) stage 2 - curate, 8:10am 10 claude code agents spawn in a sandbox. each reads today's fenced newsletters and grades them for impact. each writes a research.md. stage 3 - distill, 8:20am zoe, my openclaw, reads all the distilled research and cross-references my vault. this is where the deep personalization happens. she writes a 15-minute script. stage 4 - record, 8:25am records it with elevenlabs (audio tags are goated). then a 15-minute .mp3 lands in telegram by 8:30am. that's it. the thing is no saas can sell you this because you need to own the entire substrate and keep expanding on top of it. the meta takeaway: generation is free now. signal-to-noise gets worse every week. curated inputs shape the quality of your output. but curation itself is bottlenecked by your taste. so externalizing your thinking into markdowns enables the model to evaluate information the way you would. karpathy proposed exactly this - your personal LLM-readable wiki. it is now the leverage point that compounds forever. every note in your knowledge base is a brick in your moat.

@Shpigford honestly, that’s how I imagine you start every day.

fine. i'll launch two new businesses today.

@aaditsh why aren’t you using a tool like Wispr Flow?

I LOVE voice transcription. Probably transcribed between 500K and 1M words in the last year. ChatGPT is the best at this and it's not close. Almost no errors. And it's been this good for almost a year now which is wild. Grok is surprisingly good too. Claude is unreliable and the UX isn't great. Gemini is basically unusable. Embarrassingly bad. I use all four of these every day. ChatGPT just figured out voice transcription before everyone else and nobody has caught up.

@coreyganim @branson_atx I thought the big winners would be vibe coders who created SaaS.

implementation is absolutely where the most money is made. education is second (only really works if you have existing distribution). agree with selling an outcome, and I think speed to lead agents/automations are the best to sell because you can directly quantify the ROI also agree that LinkedIn is the lowest hanging fruit for finding clients on social (not X). so many ways to make money with AI right now.

We're overthinking AI businesses. 5 founders made $30k - $4.5MM using the same boring system. This year I interviewed 23 founders of ai companies to figure out what's working. The easiest way to win: help businesses implement ai. Some examples: + @branson_atx is a 15-year-old kid who made $30k setting up OpenClaw for people + @cheneypiano helps CEOs figure out their ai strategies and trains their teams + Joseph Zapiain went from sleeping in his car to doing over $1 MM/year setting up Zapier and other automations + @tarunthummala runs a full dev shop but companies keep hiring his agency to set up their teams with Claude Cowork + @calebhodges sets up OpenClaw in a more user-friendly way for clients + @businessbarista launched a version of this after selling Morning Brew Here's the play: 1. find hot software (like Hermes​) or result (like ai that does "speed to lead") 2. get your first clients on social Avoid X. We're already AI-nerds. LinkedIn is easier. Use DMs and "I did this" posts. 3. give clients an easy win 4. switch them to subscription Jon says his firm is his clients' "Fractional Chief AI Officer" I have an 8-page playbook based on how Jon used this approach to reach $4.5M/year. Comment "want" to get it. Or watch these interviews on YouTube.

@LouiseDSadeleer wait, they make them in different colors?

Recording my "how I got 30k views 1 month on YouTube" video today, which floofy mic should I use?

I thought vibe coders would make bank in AI. Instead, it's services.

@branson_atx How Jon does it: thenextnewthing.ai/l/jon-45millio… thenextnewthing.ai/l/jon-45millio…

@thesamparr what's what you and shaan are doing: camera facing you during the podcast gemini learns soon there will be an ai to replace you do like your mother-in-law and get into selling pillows

@thesamparr you used to post youtube links when you told us stories like this. link me

@dotta there's something about paperclip's design that's just so lovely.

David makes $2,000/mo using Paperclip 📎📎📎

@aymanalabdul @fortelabs @BearNotesApp really? Bear leaves md files on your computer?

@fortelabs @BearNotesApp is the best I've found

Can anyone recommend a dead simple, easy to use notetaking app based on markdown files? Obsidian is far too complicated for me. I'm looking for Apple Notes, but with markdown storage

@LouisShulman @fortelabs Louis, have you done anything to make your text look prettier there? I love the app but text isn't scannable there.

@fortelabs VS code

@noahkagan @noahkagan i love these updates. Ballsy

@jayclouse @rileybrown @vibecodeapp_ Baller, Jay. I'm adding it to YouTube to watch later.

He posted the FIRST TikTok about ChatGPT...and his audience grew from 0 to 200,000 in two weeks. Today, @rileybrown has 1.5 million followers, raised $9 million to build @vibecodeapp_, and one of the most impressive content machines I've seen. And he talked me through it. Timestamps: (00:00) How Riley's first ChatGPT TikTok got 20 million views (04:56) First mover advantage: (07:50) How Riley films his videos (11:24) Why structure made his content worse (13:06) My honest moment (21:33) The case for educational screen-share YouTube videos (26:45) His content strategy (30:43) The seven-account Twitter strategy that tripled revenue (37:37) Gimmicks that actually boost retention (41:00) Why AI writing your scripts is suicide in the long run (42:37) The content farm future and how to survive it (50:14) Platform rankings: where to start today

@cathrynlavery word up

@AndrewWarner Best bank ever, recommend to everyone.

When I prepped for my first interview about Mercury’s founding, I asked people on X what they thought of it. FOUNDERS RAVED After the interview, I created an account. I loved Mercury so much that I asked Immad to let me invest. So glad he said yes. It's such a pleasure to use.