正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

I recorded Claude fix two MiniJinja issues itself. It's Rust (which it sucks at), in a code base not really set up for vibe coding and it still succeeds. Yes, it takes 30 minutes, but it's hands off. Usually I let this work alone and review later.

Armin Ronacher ⇌

74,047 subscribers

39,262 次观看 • 1 年前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

10 条评论

Armin Ronacher ⇌ 的头像

Armin Ronacher ⇌1 年前

I also have the video on youtube:

Carlos DP 的头像

Carlos DP1 年前

Honestly, I’m kinda ready to say they don’t actually suck at Rust anymore haha. I’ve given them way too many problems I was sure they’d fail at and they solved it without issue. Now that they iterate with diagnostics feedback, Rust is actually ideal imo

Armin Ronacher ⇌ 的头像

Armin Ronacher ⇌1 年前

Well you can see that it struggles with tool usage and the borrow checker. Clearly skill issue ;)

ʈim anton willem 的头像

ʈim anton willem1 年前

non-stop code review fuuuck me😄🔫

Ray Quant 的头像

Ray Quant1 年前

claude's rust win shows automated fixes work, but for markets i still need proof before trusting moves.

Andrei Maxim 的头像

Andrei Maxim1 年前

This is very much my workflow! I think it works very well when you are familiar with the project and you can zoom through the diffs, but I still having a hard time relying on Claude to write new features because it’s harder for me to fully wrap my mind around the code change. I wonder if your experience with open-source projects with multiple contributors helps. I also think that the approach of discussing the bug in detail is a very useful pattern for AI-based work, because it helps the human understand what the AI should generate. I’d argue that the support for different languages and tooling will only get better over time and the quality of the code will only improve. However, it also seems that having an easy way to run a test and the entire test suite is more and more important. The same when it comes to using the right abstractions. I do wonder if the human programmer will shift towards creating the correct context for the AI agent.

Armin Ronacher ⇌ 的头像

Armin Ronacher ⇌1 年前

New features on new projects works well if you structure them for agentic workflows but it requires quite a bit of attention. I’m sure this will get better soon but today there are some things you can do to improve the success of such things. I went super deep on Claude recently and found a lot of success on new projects. After talking to @zeeg I also want to see how well Cursor’s background agents work. That might work actually better for existing projects right now because the is quite good at following rules.

Ofek Lev 的头像

Ofek Lev1 年前

What was the cost of 30 minutes?

Armin Ronacher ⇌ 的头像

Armin Ronacher ⇌1 年前

I pay a flat rate.

Michal Srb 的头像

Michal Srb1 年前

All AIs suck at Rust. I don’t think it’s the AIs’ fault.

相关视频

I did it. I figured out the perfect AI coding loop It's called the Finn Loop and it's 100x'd my vibe code output while cutting down work by 95% In this video I cover how to set it up By the end you'll have your own software factory +I completely open sourced it down below!

I did it. I figured out the perfect AI coding loop It's called the Finn Loop and it's 100x'd my vibe code output while cutting down work by 95% In this video I cover how to set it up By the end you'll have your own software factory +I completely open sourced it down below!

Alex Finn

87,765 次观看 • 1 个月前

I’ve been working on dunk. It’s a terminal diff reviewer where you mark issues inline, then let a coding agent fix them. It started as a fork of hunk, but I stripped it down to the loop I need: review → my comments → agent fixes I’m spending time reviewing code and leaving comments for agents rather than writing code myself. But in most tools, that feedback loop still feels broken. dunk is a simple way to make it work across Claude Code, Codex, Pi, etc.

I’ve been working on dunk. It’s a terminal diff reviewer where you mark issues inline, then let a coding agent fix them. It started as a fork of hunk, but I stripped it down to the loop I need: review → my comments → agent fixes I’m spending time reviewing code and leaving comments for agents rather than writing code myself. But in most tools, that feedback loop still feels broken. dunk is a simple way to make it work across Claude Code, Codex, Pi, etc.

Amir Salihefendić

13,156 次观看 • 3 个月前

I built Feynman, Claude Code for research. I gave it a question and it came back 30 minutes later with a cited meta analysis. It can also replicate experiments on Runpod, audit claims against code, and simulate peer review. Open source & MIT license, link below

I built Feynman, Claude Code for research. I gave it a question and it came back 30 minutes later with a cited meta analysis. It can also replicate experiments on Runpod, audit claims against code, and simulate peer review. Open source & MIT license, link below

Advait Paliwal

341,569 次观看 • 5 个月前

RL really sucks. It takes 10 hours just to learn breakout. ... a few years ago. It's <30 seconds on 1 GPU now in PufferLib and still dropping. Write faster code.

RL really sucks. It takes 10 hours just to learn breakout. ... a few years ago. It's <30 seconds on 1 GPU now in PufferLib and still dropping. Write faster code.

Joseph Suarez 🐡

267,591 次观看 • 10 个月前

OpenAI Agents SDK + MCP = Vibe Coding Agents I now have a CRACKED workflow to develop agents, Cursor can not only write the code but look at the logs at LangWatch, so it can debug the interaction between agents and write the fix itself

OpenAI Agents SDK + MCP = Vibe Coding Agents I now have a CRACKED workflow to develop agents, Cursor can not only write the code but look at the logs at LangWatch, so it can debug the interaction between agents and write the fix itself

Rogerio Chaves

51,352 次观看 • 1 年前

> open Claude Code > "can I edit this file?" yes > "can I read this file?" yes > click allow 30 times per session > don't even read what you're approving > there's a file that does all of this automatically > it's called settings.json > takes 2 minutes to set up > you've been clicking allow for months

> open Claude Code > "can I edit this file?" yes > "can I read this file?" yes > click allow 30 times per session > don't even read what you're approving > there's a file that does all of this automatically > it's called settings.json > takes 2 minutes to set up > you've been clicking allow for months

darkzodchi

177,856 次观看 • 3 个月前

Okay, I got Claude Code to actually fly my plane now It controls the keys via Chrome by running it with `claude --chrome` There is some issue with the lift vs gravity formula though so it takes a lot of speed to take off, but it's really cool to see it control and fly the plane This means it can code the game and then play it to check in a loop to theoretically improve it without my help (haven't gotten there though) I run it like this btw: `IS_SANDBOX=1 claude --dangerously-skip-permissions --chrome` (Run at your own risk)

Okay, I got Claude Code to actually fly my plane now It controls the keys via Chrome by running it with `claude --chrome` There is some issue with the lift vs gravity formula though so it takes a lot of speed to take off, but it's really cool to see it control and fly the plane This means it can code the game and then play it to check in a loop to theoretically improve it without my help (haven't gotten there though) I run it like this btw: `IS_SANDBOX=1 claude --dangerously-skip-permissions --chrome` (Run at your own risk)

@levelsio

143,316 次观看 • 7 个月前

There's a built-in agent inside Claude Code you can invoke just by telling Claude to use the claude-code-guide agent Save this... you'll need it It's not a local file in your project. It's a built-in subagent defined in Claude Code's system itself. It answers questions about Claude Code CLI, the Agent SDK, and the Claude API. It uses Glob, Grep, Read, WebFetch, and WebSearch, but can't edit or write files. Its only job is to answer questions.

There's a built-in agent inside Claude Code you can invoke just by telling Claude to use the claude-code-guide agent Save this... you'll need it It's not a local file in your project. It's a built-in subagent defined in Claude Code's system itself. It answers questions about Claude Code CLI, the Agent SDK, and the Claude API. It uses Glob, Grep, Read, WebFetch, and WebSearch, but can't edit or write files. Its only job is to answer questions.

Daniel San

57,240 次观看 • 5 个月前

Claude Code plays Runescape: The waterfall quest. Here Claude Opus 4.6 is acquiring quest items in the maze and tomb. Not a script, real play. It tried things, failed, learned, and succeeded. I sped this up a lot, it's ~20 minutes of real time.

Claude Code plays Runescape: The waterfall quest. Here Claude Opus 4.6 is acquiring quest items in the maze and tomb. Not a script, real play. It tried things, failed, learned, and succeeded. I sped this up a lot, it's ~20 minutes of real time.

Rob Haisfield

420,370 次观看 • 6 个月前

Stop paying for Claude Code! Use it for FREE with ollama I just set up Claude Code running locally on my computer. Zero API costs, complete privacy, and it actually works for real coding tasks. Full step-by-step guide in my latest video:

Stop paying for Claude Code! Use it for FREE with ollama I just set up Claude Code running locally on my computer. Zero API costs, complete privacy, and it actually works for real coding tasks. Full step-by-step guide in my latest video:

Mervin Praison

57,732 次观看 • 6 个月前

I found a pretty good Paul Harvey AI.... and it spit out this, after a few hours work this evening. It's not perfect, but it's what I heard in my head when I wrote this, and it's very cool to finally hear it in his voice.

I found a pretty good Paul Harvey AI.... and it spit out this, after a few hours work this evening. It's not perfect, but it's what I heard in my head when I wrote this, and it's very cool to finally hear it in his voice.

🍊🍊Capt'n Bo🍊🍊

67,187 次观看 • 1 年前

a year ago, I would have charged $1,000+ to develop this figma design now... I can do it in 30 minutes with Claude Code. not to mention, it's responsive and animated. don't believe me? watch me. where does this leave front end developers?

a year ago, I would have charged $1,000+ to develop this figma design now... I can do it in 30 minutes with Claude Code. not to mention, it's responsive and animated. don't believe me? watch me. where does this leave front end developers?

Julian Galluzzo

38,165 次观看 • 5 个月前

I tried out hunk this week and it's really good. I work in diffs now - not an editor. I need a first class way to review diffs locally and hunk showed me how to do it. But I wanted more vim in my hunk. So I put some vim in my hunk and am a happy code reviewer now.

I tried out hunk this week and it's really good. I work in diffs now - not an editor. I need a first class way to review diffs locally and hunk showed me how to do it. But I wanted more vim in my hunk. So I put some vim in my hunk and am a happy code reviewer now.

Tim Culverhouse

25,329 次观看 • 3 个月前

Ever seen a fresh (20x) Claude Max account's 5-hour usage allowance get drained in ~14 minutes? Feast your eyes on my bizarre life now with this screen recording of a recent live work session, something I've gotten at least 100 requests for over the past month. Maybe you can understand now why I need so many accounts and how I can work on so many different projects. You can also see the truth of what I was saying recently about how, once your plan is done and the beads made and polished, it's mostly just machine-tending the swarm that doesn't require much thought. Lots of just telling it to get the next bead and work on it, to review code, to re-read AGENTS dot md after a compaction, etc. And you can see how I use gemini-cli for code review. I give Google a lot of crap for the harness being broken and the capacity overloads, but when it works, it's actually really good for this code review use case. I don't usually let it write new code, though, because I think Opus and 5.2 do a better job. Also, sorry the recording is a bit blurry; I have a 5K resolution monitor and screen recordings usually are hard to watch from it. And btw, this really wasn't that normal of a session for me, it was more frenetic than usual, because I don't want to dox myself or my clients by accident. Hence all the ceaseless terminal tab swirling. I usually do more planning work while this stuff is going on, but I wanted to minimize the chances of leaking important information. That's also why I didn't refresh the Gemini login in the WezTerm window, which killed me, trust me. It's the reason I hate doing these screen recordings in the first place; it kills my productivity. Anyway, hope you liked it. I will also post to YouTube, see reply for link. Thanks for watching.

Ever seen a fresh (20x) Claude Max account's 5-hour usage allowance get drained in ~14 minutes? Feast your eyes on my bizarre life now with this screen recording of a recent live work session, something I've gotten at least 100 requests for over the past month. Maybe you can understand now why I need so many accounts and how I can work on so many different projects. You can also see the truth of what I was saying recently about how, once your plan is done and the beads made and polished, it's mostly just machine-tending the swarm that doesn't require much thought. Lots of just telling it to get the next bead and work on it, to review code, to re-read AGENTS dot md after a compaction, etc. And you can see how I use gemini-cli for code review. I give Google a lot of crap for the harness being broken and the capacity overloads, but when it works, it's actually really good for this code review use case. I don't usually let it write new code, though, because I think Opus and 5.2 do a better job. Also, sorry the recording is a bit blurry; I have a 5K resolution monitor and screen recordings usually are hard to watch from it. And btw, this really wasn't that normal of a session for me, it was more frenetic than usual, because I don't want to dox myself or my clients by accident. Hence all the ceaseless terminal tab swirling. I usually do more planning work while this stuff is going on, but I wanted to minimize the chances of leaking important information. That's also why I didn't refresh the Gemini login in the WezTerm window, which killed me, trust me. It's the reason I hate doing these screen recordings in the first place; it kills my productivity. Anyway, hope you liked it. I will also post to YouTube, see reply for link. Thanks for watching.

Jeffrey Emanuel

86,071 次观看 • 7 个月前

Klarna CEO Sebastian Siemiatkowski never learned how to code. But after a summer using Cursor, he now brings working prototypes to meetings instead of asking engineers to interpret his ideas: “Because I'm not an engineer, I've been vibe coding my whole life.” “I come into the meeting room, there's tons of product people and engineers. I say, ‘I think we should build an app that kind of does this, and it should work like that.’ They try to make sense of what is saying.” “I've never seen the code. I don't know how they made it. It's beautiful, I like our app. I think it's nice but, I don't know how it works.” “So to me, I've been vibe coding for 20 years. The only difference is now I'm sitting with a computer, and I'm doing it with AI and it's coming back with the prototype in 20 minutes.”

Klarna CEO Sebastian Siemiatkowski never learned how to code. But after a summer using Cursor, he now brings working prototypes to meetings instead of asking engineers to interpret his ideas: “Because I'm not an engineer, I've been vibe coding my whole life.” “I come into the meeting room, there's tons of product people and engineers. I say, ‘I think we should build an app that kind of does this, and it should work like that.’ They try to make sense of what is saying.” “I've never seen the code. I don't know how they made it. It's beautiful, I like our app. I think it's nice but, I don't know how it works.” “So to me, I've been vibe coding for 20 years. The only difference is now I'm sitting with a computer, and I'm doing it with AI and it's coming back with the prototype in 20 minutes.”

sourcery

149,352 次观看 • 7 个月前

It's fun to work on this animation :) Also for the weird shadow bug in the third shot.. Blender sucks at Ambient Occlusion, so I can't fix it entirely :(

Sensitive content

It's fun to work on this animation :) Also for the weird shadow bug in the third shot.. Blender sucks at Ambient Occlusion, so I can't fix it entirely :(

🏳️‍🌈~ANXI~🔞

80,429 次观看 • 1 年前

Trump Roosevelt Speech: We have the reflecting pond which we did the best job of all and they came in with box cutters and they cut it up, but it's working fine and I let a little water out. We're going to redo it a little bit, but we fixed it up. These people cut it up and they got caught. Let’s see what happens to them. A 350 ft gash. We want to have it done for July 4th, and it was done for July 4th man it's uh operating right now, but we'll fix it up very quickly Obama tried to fix it and it was a disaster.

Trump Roosevelt Speech: We have the reflecting pond which we did the best job of all and they came in with box cutters and they cut it up, but it's working fine and I let a little water out. We're going to redo it a little bit, but we fixed it up. These people cut it up and they got caught. Let’s see what happens to them. A 350 ft gash. We want to have it done for July 4th, and it was done for July 4th man it's uh operating right now, but we'll fix it up very quickly Obama tried to fix it and it was a disaster.

Acyn

67,288 次观看 • 1 个月前

I started using a tool recently too incredible not to share. Both DHH and Mitchell Hashimoto praised it and I finally checked it out. It's called Hunk and it totally changed how I review and refactor code written by/with agents. The next 5 minutes will prove it to you:

I started using a tool recently too incredible not to share. Both DHH and Mitchell Hashimoto praised it and I finally checked it out. It's called Hunk and it totally changed how I review and refactor code written by/with agents. The next 5 minutes will prove it to you:

Jilles Soeters

40,417 次观看 • 1 个月前

Anthropic CEO: “ I have engineers within anthropic who don’t write any code, they just let Claude write the code and they edit it and look it over” “At anthropic writing code means designing the next version of Claude it self, so we essentially have Claude designing the next version of Claude itself, not completely but most of it”. In the last 52 days, the Claude team dropped 50+ major feature launches. This is literally INSANE.

Anthropic CEO: “ I have engineers within anthropic who don’t write any code, they just let Claude write the code and they edit it and look it over” “At anthropic writing code means designing the next version of Claude it self, so we essentially have Claude designing the next version of Claude itself, not completely but most of it”. In the last 52 days, the Claude team dropped 50+ major feature launches. This is literally INSANE.

CG

2,268,849 次观看 • 4 个月前