Mposter

@mposter.bsky.social

Bay Area Resident 🌁 Lemon Tree Father 🍋 Real Human Engineer 🛠️ The best expectations are no expectations. Come as you are.

Everyone’s making a vibe coded agent harness with their own opinions baked in. LLMs are so good that everyone saw openclaw and decided to make their own version instead of contributing to the original, myself included.

My Vestry app was using GitHub copilot. They’ve changed their tos and now frontier models will cost a small fortune to utilize. One implementation request now costing upwards of $30.

I named my OpenClaw copy project to Vestry. The goal of the project is to take the learnings from OpenClaw and to package that experience into an easy to install/configure headless AI runtime capable of operating in home or business applications.

I drove all the way to the frontier tower in San Francisco to join an AI with Rust meetup only for the event organizer to now show up. I was looking forward to finally discussing my project and collaborating all week, and now I’m just here with my thoughts.

Anyone use GitHub copilot for LLM completions? Seems like they’ll be using our inputs as training data soon. They’re using it to train “their models” which is interesting to me because I always thought of copilot as a model proxy/provider. github.blog/news-insight...

Updates to GitHub Copilot interaction data usage policy

From April 24 onward, interaction data from Copilot Free, Pro, and Pro+ users will be used to train and improve our AI models unless they opt out.

github.blog

Always remember to create an Arc pointer to your AppState. For some reason agents don't think it's necessary, so if your vibegineered app has weird state behavior, that's probably why.

Just discovered the concept of planning markdowns with LLMs and… wow 🤯🔥 It feels like I’m not spending a lot more time refining and prompting to create the perfect planning document. My goal is to have an agent one-shot any implementation based purely on the plan doc I give it.

This OpenClaw software is 100% the future of AI agents. Not the software itself, but the concept of taking an agent, building a companion and then letting it have complete reigns of a sandbox environment. I’m sure major AI players will be releasing products around this concept soon enough.

People should run more on prem compute at home. There’s an abundance of open source software available that offer the same capabilities of many SaaS products. The barrier to using these tools is often knowledge. However with AI, I’m surprised not to see a larger movement in open source.

I had an interview today with an engineering manager of a fairly large org. An interesting topic surfaced during our discussion, the relevance of top tier communication and writing skills. Something I notice is that doc culture is championed often by org leaders, but same orgs struggle to adopt.