1000k tokens in CC is basically an approach to give you a bit more time to save your session state slightly beyond 200k. Go too far beyond that and you'll experience doom.

#claudecode

The problem with getting large language models to do what I want is that what I want is encoded in the words I use, and more importantly in the words I don't use. I don't use them because I failed to understand the problem clearly enough, I don't have enough domain expertise, or a variety of other reasons: assumptions, blind spots, laziness.

This isn't a prompt engineering problem — you can't prompt your way out of blind spots you don't know you have. It's not a context engineering problem either, and it's not just an Explore agent gathering more information for the model to read.

I've been experimenting with a dedicated agent step that takes my vague request, reads whatever context exists, and does the work of figuring out what I left unsaid. It produces something new — a synthesized version of what I actually meant, not just more context to read. Instead of the model interviewing me, it does that investigative work itself. I first called this an expander agent — it takes a compressed intent and expands it as though you had spent the time writing a detailed specification for whatever you want done.

I didn't invent this. But discovering it after trying to micromanage models, getting interviewed by them because I can't let go, and finally letting fatigue take over — that path led me here. I think it's useful enough that I wanted everyone else to know about it. The academic term seems to be intent decomposition, though what I'm doing is closer to intent expansion than the classical definition.

A bit more details about it here: https://labs.voidwire.info/posts/the-words-i-dont-use/

#llm #genai #claude

In the not-too-distant future when we're all using AI assistants — or is that now? — I wonder what happens to all the web services we've come to rely on. I don't think they go away. But I think the relationship changes. Instead of visiting sites, we'll integrate them. The services that get this — the ones that make it easy with clean APIs or CLIs — those will thrive. Not because we don't value what they've built, but because we want it woven into our systems, not siloed in a browser tab.

#ai #aiassistants #genai

So, yeah, we're all figuring this out -- don't have your LLM do things a simple script can do repeatedly and reliably. Run the script (have the LLM do it even) and return the data to it. Why? We're all becoming context hoarders -- I know I am. The more I can save the more impressed I am with myself. So, instead of taking the easy, and possibly expensive way out, think about your workflows and figure out where you can optimize. Then have your LLM write your scripts ONCE so you can context hoard for days. Welcome.

#llm #genai #claudecode #ai

I have some, probably, obvious advice when you're building with Claude Code or other agents. Don't mention MVP or prototype! It seems to induce the model to cut more corners than it already does.

#claudecode #Anthropic #llm #ai

My app is *finally* live! A local-ish only RSS/Youtube/Reddit curator that uses AI to prioritize, summarize, and do stuff that matters to YOU. You have all the data, you have all the control. Has a local API (MCP coming soon), config, TUI and some features I use everyday. Ships WITH Fabric integration!

Check it out: https://prismis.io/

#llm #claudecode #vibecoding #tui

Just released lspeak (a cli/daemon that does text to speech..includes caching and semantic searching for audio reuse). It's little project I've been using with CC for about a month or so. It's the foundation layer for another project called Clarvis (Jarvis talking to you in CC!) to be released soon. Check it out here: https://github.com/nickpending/lspeak

#llm #claudecode #vibecoding #tts #anthropic

I recently wrote about this and this continuum is how I think about the AI agent space. The further right you go, the more you're fighting against what AI naturally does well. I'm trying to stay in that sweet spot around bounded execution and bounded autonomy for some project. For others, it's WAAAY left..and AI is great at this.

#ai #genai #llm #claude #claudecode

I like that I can my various "auditor" subagents in CC to tattle on the main thread. It works surprisingly well.

#claude #claudecode #vibecoding