To dev is to be humane @dev_humane

@embirico Great analogy! It makes me think of "autonomous slider". Thanks for sharing!

@VaughnVernon @claudeai please do! Really curious what solution Claude came up with

@vibe_terminal what is this thing?

@steipete thanks for sharing! really helps me a lot!

bagerbach.com/blog/using-ai-…

lovable.dev/invite/18c624a…

@skastr052 > It allows testing "what would this look like" scenarios easily. just want to chime in here, this is also what I see the best value of this "agentic coding" thing. I start to use agent to explore ideas that would not write myself due to too time consuming or "boring" to write.

Is this setup for everyone? Does this work? Isn't this just giving you a false sense of productivity? Won't this result in complete slop code? All fair questions. I believe this is quite advanced and can potentially create a serious attention residue problem as you open several different threads but don't quite close them. If you can iterate quickly using a single thread, then I'd approach carefully, but with an open mind. Also, I understand that for this to be effective, your codebase requires an in-depth support for coding agents, beginning with the agent md files, but also extensive test suites, type-safety, thorough architecture guidelines and linter configurations. Even then, the agents are at a point where they may just completely fail occasionally, so you also want to review and discard fast, which requires experience. In this setups case, you just discard the worktree. Done. You can absolutely use this to write bad code, but why would you? Generated code is still your responsibility. Scale up as you can manage. Sometimes a single neovim window, a paper notebook, and a few hours of deep reading is the most valuable thing you can do, too. Be smart. If however you're finding yourself bored as you fire off your agent, waiting for it to do things, then the feeling of managing 10 agents across 5 tabs is like no other, and this has allowed me to do some crazy stuff so far. I'd say it's worth trying.

@skastr052 @AmpCode many thanks for sharing your setup, really helps me a lot! just curious have you tried @augmentcode ? It is an extension, and recently just released CLI. I undersand you are terminal first, but maybe you have tried it with vscode/cursor befor?

I think this is mostly right. - LLMs created a whole new layer of abstraction and profession. - I've so far called this role "Prompt Engineer" but agree it is misleading. It's not just prompting alone, there's a lot of glue code/infra around it. Maybe "AI Engineer" is ~usable, though it takes something a bit too specific and makes it a bit too broad. - ML people train algorithms/networks, usually from scratch, usually at lower capability. - LLM training is becoming sufficently different from ML because of its systems-heavy workloads, and is also splitting off into a new kind of role, focused on very large scale training of transformers on supercomputers. - In numbers, there's probably going to be significantly more AI Engineers than there are ML engineers / LLM engineers. - One can be quite successful in this role without ever training anything. - I don't fully follow the Software 1.0/2.0 framing. Software 3.0 (imo ~prompting LLMs) is amusing because prompts are human-designed "code", but in English, and interpreted by an LLM (itself now a Software 2.0 artifact). AI Engineers simultaneously program in all 3 paradigms. It's a bit 😵‍💫

swyx @swyx

3 years ago

🆕 Essay: The Rise of the AI Engineer latent.space/p/ai-engineer Keeping up on AI is becoming a full time job. Let's get together and define it.

56 361 2K 2.5M 2K

best languages for vibe coding as a tier list ask me anything

geoff @GeoffreyHuntley

a year ago

god damn zig is good i now actively recommend against rust for vibe coding the speed of the wheel matters llms hallucinate and if the compilation speed is slow the speed of generation (attempts per minute) suffers rust compilation is too slow thus hallucinations punish

13 3 125 88K 31

[👇 link below 👇] When CURSED ships, understand that Ralph built it. What comes next, technique-wise, won’t be Ralph. I maintain firmly that if models and tools remained as they are now, we are in post-AGI territory. All you need are tokens; these models yearn for tokens, throw tokens at them, and you have primitives to automate software development if you take the right approaches…

prompt (using @cursor_ai ) "see failed tests from changes in {{file}} - run yarn test -then plan how to efficiently solve all tests failures with minimal roundtrips solve all

@jackfriks Good post from @GeoffreyHuntley on the "Ralph Wiggum" hands-off-the-wheel agent prompting style: ghuntley.com/ralph/

❌ Tasklist isn’t for everything. Skip it when: – You’re debugging – You don’t know the goal yet – It’s a one-off But once the task has shape? Structure makes everything easier.

@justine_chang39 @AmpCode Just curious @justine_chang39 , any reason you prefer @AmpCode over @claude_code and @augmentcode ? AFAIK, these three have its own strength and weakness. I am now trying to figure out what is the best use case for each one of them.

@justine_chang39 @AmpCode Thanks for this great thread! I especially love the part that we should deliberately ask Amp to divide into many many many phases, so that we can check its action promptly. And indeed automation or speed is not the main goal of agentic coding, instead, quality is the goal.

How I use @AmpCode to implement entire features with minimal mistakes. A thread.

Serious question: What do you do while you wait for Claude Code to finish?

Learning out loud, what a beautiful word! View twitter/x as Knowledge building community ! #learningoutloud #knowledgebuildingcommunity