Anthony Franco

@itsolelehmann I’ve done something similar by defining “elegance” and then asking Claude if what it built meets our elegance standard “Minimally viable” imho is too prescriptive - you want maximally effective with minimal instruction

one of the highest leverage ideas in AI right now: "minimum viable prompting" the reason: your Claude prompts/skills are probably way too detailed, and its making your outputs worse boris cherny, the guy who created claude code, talks about this all the time. his own setup is surprisingly minimal way less than you'd expect from the person who literally built the tool his rule: before adding any instruction, ask "could claude figure this out on its own?" if yes, don't add it most people do the opposite. something goes wrong so they add more instructions. so the prompt gets longer. then claude follows each one less reliably. so they add more. it compounds in the wrong direction the fix: write less. be specific about the few things that actually matter and trust the model on the rest here's a prompt that identifies and cuts all the unnecessary dead weight for you. open cowork or claude code and paste this: —— i want to trim my setup down to the minimum viable instructions. go through everything: claude .md, every skill in my skills folder, every file in my context folder, everything you can find. for each instruction you find, simulate deleting it. would my output on a typical task be noticeably different without it? if no, flag it. tell me what it says, where it is, and why it's dead weight. —— also run this before you save any new instructions you'll probably lose half the words and get noticeably better results

@ErickSky @Scobleizer I’ve tried many variations of this - if your repo is well organized and documented, adding this type of rag to your codebase will degrade performance, not enhance it. The llm still needs to read the files it is touching - Give your code structure and docs to traverse

🚨 El secreto sucio de Claude Code, Cursor, Windsurf y TODOS los AI coding agents: Cada nueva sesión empiezan completamente en blanco. Olvidan tus convenciones, decisiones, arquitectura y lo que se rompió ayer… y tú terminas pegando walls of text gigantes que se desactualizan y te queman miles de tokens. MEX de @daksh_jaitly lo soluciona de una vez. Es un CLI + scaffold open source que crea un skill graph vivo de tu codebase (con ROUTER.md inteligente + archivos conectados por YAML), detecta drift automáticamente, se corrige solo y mantiene la memoria perfecta del agente. Menos gasto de tokens. Más consistencia. Cero dolor de cabeza. REPOOO👇

@adisingh It’s been around for a while - OneReach

Who’s building OpenClaw for enterprises?? Been following this space closely. Think there’s a big winner that will be born here.

@burkov Ask Claude “how do you feel about this plan?” The ask “what would make you more confident” Then decide what to do

When Claude shows me a three-page-long plan for a fix, I have two options: 1. Read it, confirm that it's all correct, and have it implemented wrong anyway. 2. Not read it, confirm it's all correct, and have it implemented wrong anyway. I prefer to choose option 2, then test, and then tell Claude that it implemented it wrong. How about you?

@asaio87 It’s not the model, it’s the harness

How are vibecoders fixing bugs ? Claude code seems to break a feature when you want to fix another. Vibe coding is a failed movement. Don’t say is a skill issue, it’s presumed I’m a non coder. Damn, it’s really made to sell tokens.

I just unlocked the Claude code max usage issues The default is to use opus at 1m context window Opus past 300k tokens is not useful (drifts) so there’s no reason to use it Change to turn off 1m and go with 200k, let it auto compact if needed Better results PLUS dramatic decrease in Claude’s usage of limits

@altryne Yep - me too - imho Anthropic did not “double usage for off peak hours” they halved usage for peak hours

My feed is showing me a bunch of folks who tapped out their whole usage limits on Mon/Tue. Is this your experience? Please comment, I want to understand how widespread this is

@svpino Define good code

Yes, these agents are absolutely horrible at writing decent-quality code. But any time you say that out loud, there's a swarm of AI-apologists who've never built anything that scream at you, "skill issue!" You might not care about the code, and that's alright, but you can't gaslight anyone who cares about good code.

Maths not mathing - even if this were true, you’re saying vibe coders are 5X more productive than a regular developer

@Suryanshti777 I went down this rabbit hole for days - graph and / or vector searches really can not touch a well structured repo This will work great for vibe coders though

🤯Holy shit....Someone just built the missing layer for AI coding. Most AI agents don’t actually understand your codebase. They grep files, open random folders, and guess architecture. This repo fixes that. It indexes your entire codebase and gives AI real semantic understanding. Not just files — everything: • dependencies • architecture • APIs • infra configs • database schemas • cross-file relationships Then I saw the benchmark… Tested on the VS Code codebase (2.45M lines): → 61.5% less context used → 84% fewer tool calls → 37x faster than grep-based agents That’s not incremental. That’s a completely different workflow. Instead of AI searching blindly, it actually knows your codebase. Zero config. Fully local. Works with Claude Code, Cursor, MCP, Codex. This might become standard for AI dev tooling. Repo in comments ↓

@matthewmillerai Claude Code cli opus 4.6 high thinking for code, architecture, and planning (but NOT using the default Claude plan- it sucks) Codex 5.4 for code reviews ONLY (will catch bugs and issues Claude misses)

Claude Code is superior. Period.

@levelsio Yes - it’s a setting in Claude code cli - just paste the message in Claude code cli and ask it to ipdate its own setting - you’ll have to restart It took me twice but it’ll get it right

Any way to skip this question when I run Claude Code in / root?

@clairevo I agree - these features feel totally disconnected from one another - and if you choose to rely on them and build your processes around them you’re signing up for a mountain of tech debt

I’m not gonna bet against Claude Code, but there is a very important lesson in product competition going on right now, and it’s this: The features ≠ the Product

@WizLikeWizard Use opus or sonnet and it’ll change your perspective Also - install Claude code on the same machine - point it to openclaw - tell it yo understand it and write a Claude.md file in it for future use, and now you have an openclaw assistant

Have been using OpenClaw for ~a month and it kinda sucks? I spend more time battling it to get basic crons fired reliably, remember things, and not repeat itself. Am I doing it wrong or are we just still very early on all of this?

@Palanikannan_M Iterm2 w/tmux

pov: you tried cmux and realized tmux was right there the whole time all i needed was "is my agent done?" and fast switching. not a new app. runs inside ghostty. same tmux shortcuts. freakishly fast

@sweatystartup Or things will level out once the cost of inference drops

People can’t handle the truth. It isn’t sustainable. Switching costs are virtually nothing. The bubble will pop when the VC subsidies stop.

Every time I try and use something other than opus 4.6 - I’m severely let down

@sweatystartup Taking a picture of this post to remind nick in 2 years about it Won’t age well

Ford shutting down Electric Vehicle production. Meta shutting down Virtual Reality investment. Only a matter of time until companies pull way back on Artificial Intelligence.

@rohanpaul_ai this is click-bait Calculators harm our ability to do long-form division.... you could say the same about that disruptive technological innovation as well...

Anthropic's own study proves Vibe-Coding and AI coding assistants harm skill building. "AI use impairs conceptual understanding, code reading, and debugging abilities, without delivering significant efficiency gains on average" Developers learning 1 new Python library scored 17% lower on tests when using AI. Delegating code generation to AI stops you from actually understanding the software. Using AI did not make the programmers statistically faster at completing tasks. Participants wasted time writing prompts instead of actually coding. Scores crashed below 40% when developers let AI write everything. Developers who only asked AI for simple concepts scored above 65%. Managers should not pressure engineers to use AI for endless productivity. Forcing top speed means workers lose the ability to debug systems later. ---- Paper Link – arxiv. org/abs/2601.20245 Paper Title: "How AI Impacts Skill Formation"

Anthropic: Double usage for the next 2 weeks! Claude: {Error 500 - Service is Down}