I've been thinking about how, in software engineering especially, it feels like the balance of who is the builder and who is the user is shifting. It used to be the case that the software engineer...
Spent several hours coding with Cursor's composer-1 model. I like it a lot and it is so fast. When you have it working it's one of my favorite agent coding tools I've used. Well done.
I've fallen behind on blog things. In the past few weeks, I've been experimenting with claude-agent-sdk, equipping the agent with MCP and prompting to use a CLI measure trade-offs and...
I built a Claude Code slash command that given a prompt, analyzes the ongoing chat and captures to insight or learning from the conversations. This seems like it's going to be useful for capturing...
I learned about the TOON input data format for language models. I wonder if this could be useful at all for the prompt built by DSPy.
I've been experimenting using Claude Code to learn Rust.
Cursor released their "Composer" model today. I believe the name is a hat tip to the original name of their agent/multi-file editing mode.
I played around with the idea of using a language model as a sort of generic response engine for any request using any protocol. That plan was to write a system prompt that the language model would...
I've now run into the following issue multiple times using Bedrock. It's my own mistake but I end up confused each time.
I've been finding that pairing with someone to help them get familiar with building with agents to be a very effective way to help them get started successfully. Everyone enters the process with a...