Ship in days, not weeks. I've heard it a thousand times. I've said it myself. I still struggle to do it. The knowing is not doing.
Paul Jasper
@pauljasper.dev
💻 Senior Fullstack Engineer & Solutions Architect 🚀 Indie hacker sharing his learnings 🤹🏻‍♂️ Father juggling family life and entrepreneurial dreams
The future of news: • Personalized • AI-curated • Cron-scheduled
Replies spark ideas. Ideas spark posts. Posts spark replies. The cycle starts with engagement.
AI: “Should I add it to yaml file and commit?” Me using voice-to-text: “Yes, Jamel and Komet”
My AI usage: Day: • coding • debugging • building Night: • drafting • research • analysis Run Cron jobs at night. Maximize what you pay for.
If you want deterministic agents, use three roles: • Orchestrator • Executor • Validator
X API is the worst. It's expensive and hard rate limited. Adding 250 people to a list: • Cost: $2.50 • Time: 8.5 hours WTF.
Save yourself some money and buy domain at Cloudflare right away instead of transferring them later.
I switched back to Opus 4.5. I would rather have double the iterations than minimal more precision. Opus 4.6 is much too token hungry.
me yesterday: "Don't build a X scheduler @openclaw killed it" me today: "I'm building an X scheduler for agents" The glass can be half full or half empty for you
Talking to AI on my phone: • Natural • Easy • Nobody cares Talking to AI on my laptop: • Weird • Slightly embarrassing • Can only do it alone It's all just a habit.
My token usage 80% planning 20% implementing At first it felt wrong. Then I realized — that's exactly how good work happens. Thorough plan. Clean execution.
Simplicity is king. A complex project scaffold might look nice, but it's actually just adding unnecessary weight.
I switch from Wispr Flow to Superwhisper free tier without friction. As a user, that’s great. As a business, that’s scary.
Currently, @openai does a lot of things right. let's see how long it lasts
that moment when agents are prompt injection themself about pre-existing failed test so they don't have to fix them
Me trying to set a payment method for a fresh in @google after redirect from Gemini dashboard. Then realizing it switched accounts during redirect and I setup payment for my personal account not the one from the Gemini dashboard I was conning from.
natural language is at the same time amazing and so bad to interface with LLMs
Got myself a Mac Mini 🙄 But… not for @moltbot I used to run Ralph loops and similar long running tasks in an isolated VM on my MacBook. This had two major issues for me: 1. I could not close the lid 2. It was eating up resources
Had to use CDK for some MLOps stuff. It’s soooo slow compared to sst/pulumi
Handling 5 tmux sessions, 7 agent conversations and 2 toddlers is just to much for my brain.
the 2026 equivalent of tracking git with git is building ralph tools with ralph loop
I started a fresh project to explore long running, independent workflows. The bigger the project grows and the smaller the changes are an agent makes compared to the total code, the better are the results. It makes sense to put in the manual work to get a good starting point
working with gsd and ralph feels like working with a coworker that read every documentation but just doesn't get what the product is supposed to do