Just created a professional investor deck in 15 minutes with @gamma No design skills needed just typed what I wanted and AI built: Problem/Solution slides Market analysis with graphics Financial projections Team profiles Competition matrix From idea to shareable deck in minutes

Most people don’t want to be sommeliers they just want to order wine like they’ve got it all figured out so we made SOMMIA the chillest AI sidekick for that exact vibe

Me vs the free-tier AI quota wall Trying to jam a startup-sized idea into the tiniest compute limits It’s like pitching a unicorn to a goldfish bowl End result: a lot of splashing, not much swimming Anyone else hit this peg-not-fitting-hole energy lately?

Day ≈ 80 247/928 1% at time

Pieter's tweet got me thinking (@levelsio ) If you don't have an app almost finish or already up and running Make a wrapper with Nano Banana 🍌🍌🍌 Here are the two most relevant ideas I found: 1. StickerFuse The pocket sticker studio Killer Feature: Photo-to-Sticker. Instantly

Day ≈ 82 251/936 Lesson of the week Obsessed beat lazy every time Keep sharing your journey

I saw this on the internet and thought it was worth sharing

How I paid for my Dev Apple Access this Sunday A guy I know told me on Friday that a web agency wanted €1K to build his website He works at heights and just wants a simple showcase website, nothing complex So I spent my Sunday making him a nice website for €300, going back and

Day 85 Locked in I apply the same thing to myself If there are days when I am negative

Theory craft anthropic reduced the context window and reduced the “intelligence” of the model for over a month in order to have a nice, profitable table to show to the VC, and we should find Claude Code back on top within a week 🙏

My Plan With Claude My MVP

That SommIA effect Saved a dinner convo from the dreaded awkward silence 10s to the perfect wine match No sommelier badge required

writing AI code is like playing slots hope with every token frustration with every bug except the jackpot is a passing test suite

This may be the blueprint for building practical AI tools fast: 1) Wrap ChatGPT/Claude into apps builders actually want 2) Gather real user feedback and track usage patterns 3) Fine-tune open models on that data and RL on feedback to match ChatGPT/Claude Same quality, more

"Technically fixed" is dev code for "it runs on my machine, not touching it again"

Sending code straight to prod is a team sport now AI already dropped 30 points off the bench Meanwhile I’m still reading the rulebook

Every dev has a secret code for "works on my laptop, not touching it again"

Sommia is everywhere