More from @trq212
Sep 22
Your Agent should use a File System
This is a hill I will die on. Every agent can use a file system.
The file system is an elegant way of representing state that your agent could read into context & allowing it to verify its work.
🧵on why and examples
Claude Code was what proved this to me.
Before CC, people though context windows would get long enough that you could just fit a codebase into context.
But that is not how programming works. You don’t need to remember everything, you just need to know how to find it.
Here’s an example from my open source email agent (github.com/anthropic-expe…)
Instead of dumping a ton of emails into context, I write them to a file and let the agen…
More from @trq212
Sep 22
Your Agent should use a File System
This is a hill I will die on. Every agent can use a file system.
The file system is an elegant way of representing state that your agent could read into context & allowing it to verify its work.
🧵on why and examples
Claude Code was what proved this to me.
Before CC, people though context windows would get long enough that you could just fit a codebase into context.
But that is not how programming works. You don’t need to remember everything, you just need to know how to find it.
Here’s an example from my open source email agent (github.com/anthropic-expe…)
Instead of dumping a ton of emails into context, I write them to a file and let the agent grep across those files.
Fundamentally this works because it lets your agent have multiple passes at a problem and let it fix its work.
In this case it can try a few different address searches, correlate them to exact lines to make sure it doesn’t hallucinate and extract them in a structured way.
Read 6 tweets
Sep 17
Making an Email Agent using the Claude Code SDK
If I wasn’t at Anthropic, I would be making agents using the Claude Code SDK.
But doing > talking. So I’m building in public and open sourcing a local email agent.
This is part one on agentic search.
First, why the Claude Code SDK?
Whenever I’ve built an agent in the past I ended up hand rolling the same patterns that are now just way easier to use the Claude Code SDK.
In this example I’m using it for subagents, context management, the file system & code generation.
Building Context
Email has a ton of context on me. My ideal agent should be able to (with permission):
- Know info about me like my address, phone number, etc
- Write intros based on the context I have with contacts
- Find all relevant emails to draft a response
Read 7 tweets
Aug 7
It’s maddening that we’re creating real time AI game engines, but there aren’t any actual AI games.
🧵 here’s my take as an ex-gaming founder on why
People don’t want generative slop.
The edge that games have over other media is achievement.
As Tiktok increasingly offers the best ratio of effort to dopamine, games need to lean into achievement and mastery.
In an infinitely generated world, where do you find the mastery?
Creation is Fun
Is Claude Code an AI video game? I’ve had several people make that claim to me.
It’s easy to learn and hard to master. It rewards you with joyful output.
It feels more like riding a horse than turning on an oven.
Read 6 tweets
Jul 22
Use Claude Code as your Video Editor
Believe it or not, every video you see below was vibe coded by Claude Code.
🧵 here’s how you can do it too:
The UI videos are all powered by Remotion ()- a library for creating videos using React.
And Claude Code is great at writing code for Remotion! remotion.dev
Just take your UI library components, pop them into your remotion codebase and ask Claude Code to make compositions that use them.
An additional benefit of making videos with Remotion is that they can easily be scaled to different screen sizes, copy can be changed and internationalized quickly.
Read 6 tweets
Jul 14
Claude Code is All You Need
When I first joined Anthropic I was surprised to learn that lots of the team used Claude Code as a general agent, not just for code.
I’ve since become a convert! I use Claude Code to help me with almost all the work I do now, here’s how:
Why?
In Claude Code Everything is a File, and it knows how to use your computer like you do.
Name your files well, and CC will be able to search them like you would.
This lets you make custom setups for memory, todos, journals, screenshots and more.
Here’s, my setup:
I use Mac OSX and run Claude Code in my home directory (~).
I have a that tells it how to access important directories in my folder.
I have different folders for memory, journals, ideas, code, to dos, memes, and scripts. Claude.MD
Read 8 tweets
Mar 13
✨ New AI Interfaces powered by Interpretability
I’m excited to share LatentLit, the result of my applied AI research fellowship with @GoodfireAI
Mechanistic interpretability isn’t just important for AI safety, it also gives us new ways to steer and interact with LLMs.
In LatentLit, you write stories by like a DJ might make music, adjusting knobs and dials using steering and seeing what effect they have. You might call it Vibe Writing.
Try it out here!
Our neural map highlights features like “Supernatural Discovery” and connects them to specific text passages they influence.
Each colored bubble in the neural map represents a feature the model activates. The bigger the bubble, the stronger the influence.
Hovering over these features will highlight where in the text this feature was activated most strongly.
Read 4 tweets