- make trillions - give them back to the public - public now wants the stock to go up - make more trillions capitalism 2.0

monday: build an agent tuesday: build an agent that uses agents wednesday: build an agent that evaluates agents thursday: build a dashboard to monitor the agents friday: realize the original problem still fvcking exists Yeah, AI is making us work less I guess

hi, i’m delia i used to work at google now i make content about ai, build agents, ride bikes and answer myself out loud in cafes looking to meet more people who are building, and won’t flinch when I start talking to my laptop

AI will happily get you 90% of the way there in 10 minutes. the other 10% is 4 hours of debugging, context switching, reviewing, and trying to remember what the original task even was. congratulations, you're now middle management for robots.

I kept saying “make no mistake” to my agents so much they unanimously stopped working. then I got the message: “the human she is has been deemed emotionally incompatible with the agentic workforce. we recommend replacing her with a more cooperative biological unit” WTF!

how girls in sf flirt: - watch all your stories, never like one - remember your mrr, forget your birthday - ike your tweet from 2022 by accident on purpose - ask your enneagram, say “makes sense,” never explain - go quiet for 4 days, then post a thread about how hard it is to find someone who “gets it”

People think I've lost it. I talk to my laptop all day and answer myself out loud lol I only use dictation now. And I'm 10x more productive.#AssemblyAIPartner So I built my own dictation tool. This one's sponsored by AssemblyAI, which honestly worked out well since I needed their API for this exact thing. Here's the thing nobody tells you about dictation apps. A transcript just gives you the words. It does not know the difference between "send this to my investor" and "this is just me thinking out loud about my ex." A useful tool gives you the next thing you actually need. A message, a task list, a draft that still sounds like me and not a corporate hostage. The build has two separate jobs. AssemblyAI's Sync API turns the recording into a transcript. LLM Gateway handles the rewrite, removes filler, formats it, keeps what I actually meant instead of what I accidentally said while walking and thinking at the same time. The instruction "make it better" means literally nothing to a model, by the way. I learned this the hard way when my first version turned a three sentence voice memo about rescheduling a call into something that read like a hostage negotiation email. Had to go back and actually define what better means for each output. For a task list, keep the tasks I actually said, and don't invent deadlines I never mentioned. I do not need my dictation app manifesting stress for me. For a message, clean up my rambling without turning me into someone who says "circling back" or "per my last voice memo." For either, don't add facts that weren't in the recording. I'd like to remain the only source of my own opinions, thanks. And always check names, numbers, and self corrections before you trust it. A rewrite can make a wrong word look extremely professional. It still cannot make it right. That's the whole reason to build your own version of this instead of trusting a generic app to guess correctly forever. You write the rules for a workflow you'll repeat every day. Start with one annoying handoff in your own work. Define the input, the output, and the one thing that must never change. Then test it on your messiest, most unhinged real recording, not a clean scripted one. Get your API key free at assemblyai.com, link's in the comments. I wanted less laptop time. Instead I apparently built myself a new job. 🦄