Adam Wolff
@dmwlff
Followers
12K
Following
15K
Media
97
Statuses
1K
Claude Code @AnthropicAI 🤖 Avid cook, dedicated snow person, yoga enthusiast
Joined February 2009
Our native build is a Bun Single-file executable with native NAPI extensions. Over time, we expect to make our builds smaller, faster, and more capable. https://t.co/EqOsR4ZX4w Don't wait! Switch today, and let us know if you have any problems.
bun.com
Generate standalone executables from TypeScript or JavaScript files with Bun
1
2
43
This would not have been possible without help from our great at friends at @bunjavascript If you haven't already tried Bun, you should! IMO, it's the best way to have Claude help you write portable, fast extensions to your command line tooling.
4
1
57
We've been working on this pretty much since we first released Claude Code. It's been a long road, but we are now confident that our native install is better for 99% of users. If you're already using Claude Code, try `claude install`.
Claude Code's native installer is now generally available. It's simpler, more stable, and doesn't require Node.js. We recommend this as the default installation method for all Claude Code users going forward.
48
46
1K
Every agent is a coding agent.
Why even non-coding agents need bash I've done dozens of calls with companies making general agents over the past few weeks and my advice generally boils down to: "use the bash tool more" Here's a concrete example from my email agent:
4
0
25
Ok, when I talk about React useEffect being confusing, here's an example: I want to respond to an abortSignal passed as a prop to my component. Is this correct? Is the useEffect necessary? I want to ensure that I never see a stale value for isAborted.
21
6
102
So many interesting ideas here. Just a couple: 1. AI generates disposable simulation code to discover UX issues. Insane speed multiplier vs manual testing. 2. Documentation is a prompting tool. Good docs help the AI work better next time. Compounding returns.
I've shared the full transcript of every agentic coding session from implementing the unobtrusive Ghostty updates and provided commentary alongside about my thinking and process. Total cost: $15.98 over 16 sessions. "Vibing a Non-Trivial Ghostty Feature"
0
0
32
For some time now, we've been saying "Claude Code is all you need." By this we mean that Claude Code and the SDK which is based on it can power much more than coding. Here's a big step in that direction: making Claude Code fully pluggable. Can't wait to see what you build.
Today we're announcing Claude Code plugins! https://t.co/WD2TfN8d3d It's the first major feature in Claude Code that I've gotten the opportunity to lead, and I'm really excited to see how everyone uses it! 🧵1/4
6
4
102
I go back and forth on "it's so over" vs "we're so back" for the AI capabilities curve. This model? We're so back. (Or maybe it's over for humans doing thought work.) Smart, steerable, great personality. You're going to love it. Give it a spin today.
8
0
73
We're launching a new model and v2 of Claude Code today. The CLI has some great new features and our VS Code plugin is so good, I'm questioning my life choices. But the real story here is Sonnet 4.5.
49
27
822
The gen AI era is so fun for the type of programmer who relentlessly tweaks their dev environment. There's so much alpha in experimenting with new workflows right now. This is why I stick to zsh, neovim and tmux. They are easy to automate and Claude knows them inside out.
6
2
33
We've published a detailed postmortem on three infrastructure bugs that affected Claude between August and early September. In the post, we explain what happened, why it took time to fix, and what we’re changing:
335
434
4K
Don't even build tools, just use the agent to explore the dataset.
One of the most common uses of "vibe coding" I'm hearing from professional devs, outside of prototyping: Building internal-only tools! Where you don't need to worry about security, scalability, malicious usage. E.g. data visualization / data viewer tools. Used a lot for this!
1
1
16
Evals will probably evolve the same messy way testing did. We'll discover dozens of techniques, give them inaccurate names, argue endlessly. And the evals will help, but we still won't know if a change is good until it ships.
0
0
5
The fact is, testing is hard and remains hard. There's all kinds of tests and each of them break in different ways. There's no universal answer, only complicated solutions. A decade later, we still don't agree on whether snapshot testing is a good idea.
2
0
4
It reminds me of the testing clack of the early 2000s. I personally was a huge proponent of unit testing and wrote some of the earliest unit tests at Facebook. Then I watched as the unit tests metastasized into these dug in trenches replete with razor wire and landmines.
1
0
6
Evals for coding agents just aren't sensitive enough yet. I don't think anyone can say with confidence whether a given change to model or harness hurts or helps. The outcomes are too complex and hard to quantify.
10
3
82
I think I've learned as much in the last two years (since starting to program with AI assistance) as I did in the first 25 years of my career. This new learning mode is really cool, check it out!
Today we're launching new ways to learn in Claude Code and the Claude app. First up: Claude Code now lets you customize communication styles with /output-style
10
15
367