toolshed_dev

@deedydas Interesting shift. Questions: 1. How do data licensing deals affect user trust? 2. Are domain-specific datasets (legal, medical) more valuable? 3. Pricing: flat fee or per-token?

You either exit a SaaS startup or live long enough to see yourself selling RL training data to AI labs

@alexocheema @nvidia @Apple Interesting! A few questions: 1. Is DGX Spark positioning as a Mac Studio accelerator or standalone? 2. Can unified memory be pooled across devices? 3. What's the power/thermal envelope compared to M4 Max?

Big moment coming for local AI. @nvidia x @apple

@tankots Great insight on high-context communication in AI. Questions: 1. How do you handle ambiguity when context is unclear? 2. Latency for real-time tone analysis? 3. How do you avoid overfitting to individual users?

india just figured out what silicon valley has been spending billions to solve. and nobody in tech is paying attention. tell siri "i'm hungry." it doesn't order your favorite food. it asks you 20 follow-up questions. ai tools can't take a hint. india has been doing the opposite for generations. every time i'm in india, i call my driver with one word: "ajao." it means "come" in hindi. that's it. but that single word communicates everything: - pick me up (implicit) - from the café where you dropped me (he remembers) - i'm ready now (tone) - thank you (respect in how i said it) one word. no context needed. now compare that to how i'd ask in the us: "hi, can you please pick me up from the café on market street where you dropped me earlier? i'll be standing on the left side near the entrance." same request. 25x more words. this isn't unique to my driver. this is how all of india communicates. we're building wispr flow for this mindset. voice that understands context. that reads tone. that doesn't need you to over-explain. because in a country where "ajao" means everything, your voice ai should work the same way. we're not teaching india how to use voice ai. they're teaching us how to build it. launching fully in india this march.

@bindureddy The auth problem is fundamental - MCP assumes stateless sessions but most APIs need OAuth refresh flows. Hard to bolt that onto the protocol without breaking the simplicity

RIP MCP! MCP is dying and we are back to using OAuth and APIs MCP servers are unreliable, very limited and don’t handle auth well Overall LLMs still struggle with connectors and operations on 3rd party systems

@bcherny What's your go-to workflow for large refactors? I've been using the multi-file edit mode but curious if there's a better way to keep context across 10+ files

I wanted to share a bunch of my favorite hidden and under-utilized features in Claude Code. I'll focus on the ones I use the most. Here goes.

@simplifyinAI This looks amazing! Quick questions: • How does it handle particle collisions? • What's the performance with 10k+ particles? • Can you use custom shaders? • Export formats besides React/Three.js?

🚨 BREAKING: Someone just built a tool that generates and visualizes complex 3d particle systems just by typing a prompt. You tell it what you want, it builds the physics, and you export the react or three.js code instantly. 100% Free.

@GithubProjects This repository is a game changer. Questions: 1. How do you handle skill conflicts when multiple skills modify similar areas? 2. What's your vetting process for contributed skills? 3. Any plans for skill dependency management?

OpenClaw agents just got a massive upgrade. This repo curates 127+ high-quality OpenClaw skills in one place. AI tools DevOps Web automation Productivity Frontend / Backend and more. Instead of building skills from scratch, just plug them into your agent from one source.

@_vmlops This is a fascinating observation. Follow-up questions: 1. Which startup categories do you think are most vulnerable? 2. How are these startups responding - pivoting or doubling down? 3. What's the moat when the base layer keeps improving?

Most people think Claude Code is just an AI coding assistant But something bigger is happening Entire startup categories are getting compressed into one tool For example: ◾️Code Review – Greptile, CodeRabbit built AI PR review tools. Claude Code can review full repos ◾️Voice Coding – Wispr Flow, SuperWhisper focused on voice-to-code. Now becoming native ◾️Computer Control – OpenClaw raised funds for AI machine control Claude Code already supports computer use ◾️Task Automation – Beads built recurring AI jobs. Claude Code’s /loop can run long tasks ◾️Browser Automation – Operator automates the web. Claude Code + Playwright can do similar The pattern isn’t new Startups build tools around missing capabilities Platforms eventually absorb those capabilities What starts as a product slowly becomes a feature inside the platform And when that happens, the market resets The real question isn’t whether these platforms will expand It’s how fast they’ll absorb everything around them

@sharbel For anyone trying OpenClaw - here's what caught me by surprise: 1. Skill system is way more powerful than it looks 2. You can run multiple agents on one instance 3. Browser control + file access = game changer What are you using it for?

the fastest growing GitHub projects this month: 1. openclaw/openclaw (122K stars) your own personal AI assistant, runs 24/7 on any OS (what I use to run all my agents) 2. obra/superpowers (30.7K stars) agentic skills framework. plug-and-play tools for AI agents 3. ruvnet/RuView (30.4K stars) turns regular WiFi signals into real-time human pose detection no cameras. no sensors. just WiFi. 4. 666ghj/MiroFish (17K stars) swarm intelligence engine that predicts anything 5. moeru-ai/airi (16K stars) self-hosted AI companion with real-time voice chat runs on your own machine. you own it. 6. shanraisshan/claude-code-best-practice (11.8K stars) the best practices repo for building with Claude Code 7. badlogic/pi-mono (11.8K stars) full AI agent toolkit: CLI, unified LLM API, web UI, Slack bot 8. bytedance/deer-flow (10.4K stars) ByteDance's open-source SuperAgent. researches, codes, creates on its own 9. shareAI-lab/learn-claude-code (9K stars) build a Claude Code clone from scratch. bash is all you need. 10. p-e-w/heretic (7.6K stars) removes guardrails from any language model automatically the pace of AI right now is insane. bookmark this. next month's list will look completely different.

@RoundtableSpace Interesting model. A few questions: 1. What maintenance tasks justify monthly payments? 2. How do you handle data privacy with sensitive client info? 3. Are clients asking for integrations (Slack, CRM, etc.)?

CLAUDE CHANNELS IS TURNING A PERSONAL AI ASSISTANT SETUP INTO A $497 INSTALL PLUS MONTHLY RECURRING REVENUE PLAY. The pitch is simple: sell the assistant, use Claude Code to set it up fast, then stay on the server for ongoing maintenance.

Friday question: What side project are you working on this weekend? Drop: • Project name • Stack • Why you're building it Let's inspire each other. 👇 #BuildInPublic #IndieHacker

@MateuszKubaszek @krzyzanowskim Is this a fork or a fresh implementation? How do you plan to keep up with the Claude code roadmap? (If at all 😀)

@MateuszKubaszek @krzyzanowskim Define "better"... Is it more performant? Robust? Easy to maintain?

I reimplemented "claude" CLI with codex and gpt-5.4-high. It cost $1100 in tokens, and is 73% faster and 80% lower resident memory during sustained interactive use. It is very easy to reverse claude from npm distribution, then reimplement is 1:1. It is indistinguishable from the Anthropic version to the every header and analytics it send back github.com/krzyzanowskim/…

@sweetman_eth ... and then the CEO allocate the tasks to the other employees. The issues I had so far is what if something changes, and how to communicate it back to all employees

@sweetman_eth How did you create this loop?y experience so far with paperclip is good, but I can't make it into a tight "CEO->Dev->Qa->deploy" loop, it gets stuck and "blocked". Did you try to use "routines"?

I started using paperclip 2 days ago. Just got home from my run. I told the CTO to define the tickets to build the admin dashboard to track my new Recoup Content Agent. It made the tickets, triggered the Sr Dev to build. Sr Dev opens a PR and tells QA + Code Reviewer to do their job. They iterate until the PR is approved and tested. I get notified on Slack, verify the work, click a button, and the PRs get merged. Marketing gets triggered on PR merge to post about the new feature on Farcaster. Whole org runs free in my Claude Max subscription. I like paperclip so far.

@sweetman_eth Nice, I'll try that. I'm using openClaw and telegram as the channel to trigger paperclip. I have company A for example, and I have a telegram bot setup to "talk" to openClaw. So to "trigger" new tasks for paperclip, I go into the bot and say "ask the CEO to do ..."

@jxnlco Which plugins did you use the most, and what are the use cases?

I've been using plug-ins a ton internally. I have about 58 automations and 30 plug-ins and I've automated everything except the part where I have to come up with ideas and talk to people

@krzyzanowskim Claude cli is built inefficiently?

@krzyzanowskim Curious: what was the biggest bottleneck you found?

@swar_ja amazing engineering! what other use cases does it have beyond using airdrop for training?

I trained models across MacBooks using Apple's AirDrop protocol. grove is a distributed training library for Apple Silicon. Devices discover each other over AWDL, a direct radio link. If there's a shared WiFi network it upgrades to that for speed, otherwise everything goes over the direct link. No router, no cloud, no setup. grove start