Turns out I just needed to warm up better. I just hit my personal best for distance, only stopped because it got dark out.

RT @Shpigford: pretty sure millions of javascript developers would switch to Rails if they’d just try out ActiveRecord for a week.

Use AI agents with usage based pricing, you're less likely to experience a rug-pull.

At least the pace is improving.

RT @wesbos: If you can’t center things in CSS by now, Is it’s 100% a skill issue.

How do I run without destroying my shins every session?.

MCP servers are more than just REST API counterparts: . You can nudge the model toward certain behaviors by telling it how to handle your output. With lots of tools available, this can make the choice easier for the model.

Finally got my MCP to have the simplest possible start script. Its just `npx graphsense`

RT @_svs_: Ok the new modelcontextprotocol typescript SDK makes it dead simple to build OAuth backed MCPs. Works really well. Host your MCP….

I'm building a desktop app for my code intelligence MCP. It spins up a vector DB and knowledge graph for each repo and re-scans them if any file changes. It hosts a separate MCP server locally for every repo, which you use in your coding agent for added intelligence!

Apparently every node in BEAM (Erlang VM) maintains an internal nameserver to dynamically add or remove nodes to a cluster. But it connects them via a mesh network, so you get 5000+ total connections (and heartbeats) in a 100 node cluster!. There is no free lunch.

@jdorfman I've open sourced it, I hope people find it helpful!.

I wanted to test my MCP server, so I built a Postman for it. You can even use Claude and it will call your tools when necessary. It's MIT licensed, link below:

"Adding a console.log to each function handling authentication". With semantic search: 11k tokens used.Without semantic search: 25k tokens used. MCP server coming in handy.

Coding agents are cool but they all need to do semantic search on the codebase. ripgrep is not enough.

I knew nothing about AI agents a month ago. Now I've built a code-search agent that can take decisions and retrieve context on its own. You can just do things.

Implemented proper tool calling for LLM-based decision making and context retrieval.

Felt the AGI today. I automated the boring parts at work using @zeddotdev Agent mode and it required very minimal hand holding. 10/10 would recommend.

Performing semantic search on ambiguous queries, before getting all the nodes. Turning out to be a proper RAG layer for codebases