San Francisco, CA
I'm a software engineer, designer, and the founder of Jellypod. I write about things I find interesting — practical applications of AI, product/design, & generative media tech.
Assigning issues to agents that actually create the entire PR is great, until you need to test what they wrote. Here's how we configured an end-to-end preview environment for Jellypod using Supabase, WorkOS, and Vercel to test agent-written PRs without ever testing the code locally.
It took 2 hours and $25 in tokens. When AI can create software overnight for a few dollars, why spend hundreds per month on SaaS?
I haven't opened a code editor in weeks. AI writes all my code. But my passion was never about writing code — it was about building cool things.
Software development and cooking are strangely more similar than they might first appear.
Generalized best practices exist for a reason - but how do you know if they apply in your specific situation? When should you question them and spend time reasoning if it should be applied blindly?
By redesigning the system to use fine-tuned models through OpenPipe, I reduced inference costs by 85%+ while significantly improving output quality.
The way businesses will be built going forward have the potential of unlocking trillions of dollars of new value, but businesses who fail to quickly adjust will quickly die.
A constitutional amendment to clarify US citizenship is a human-only right might seem absurd, but rapid advancements in AI and their increasingly human-like interactions raise important questions.