Runtime

Runtime was a tool that essentially let you run repetitive prompts without having to actually type the prompt every time. Generally I wanted to be able to craft a prompt on my computer, in detail - and then be able to use all that on my phone without needing to copy-paste a prompt.

I built it to include a Discovery element too, I was hoping I’d get some traction from people publishing their Runtimes.

I wrote about building Runtime since it was basically the beginnging of “vibe coding”. So I leaned into that!

What worked

• Learned a lot about LLMs and prompting, chaining those prompts.
• Got familiar with Supabase, some pros/limitations of Edge Functions - heavily utilised here
• Lots of attention to the project, thousands of users, thousands of Runs

What didn’t work

• Short term - OpenAI dropped GPTs the following week and it was game over.
• Long term - LLMs got smarter, developed memories, my usage patterns changed. The simple things I wanted to accomplish the LLM can do natively; the more complex things are better served by fine-tunes or MCP servers.

Generally, the most fun part of this was @rauchg (Vercel CEO) tweeting about my project.