Hunt3rr 

@kelvinbuildss no doubt

Hot take: No matter how good you are, If you're still coding everything manually, you're wasting everyone's time. do you agree?

@xikhar we need thattttt @claudeai

Anthropic could never match this

@sarveshhon let's do it!

Let’s connect If anyone🚀

@anupamrjp and don't even start with grok

think GPT-5.5 or Opus 4.8 is disappointing? spend 5 minutes with Gemini.

@1thousandfaces_ claude code doesnt need a splash screen, the vibe is just a terminal quietly judging your git history

no image gen models on the market rn could make a global village coffeehouse loading screen for claude code to contrast with frutiger aero codex...sigh

@goyalshaliniuk spec-driven is the right instinct but the spec is only as good as the context behind it. watched spec kit spit out a beautiful plan off one wrong assumption and confidently build the wrong thing. garbage in still applies

GitHub just fixed the most frustrating part of vibe coding. They released a free tool called Spec Kit—and it’s already crossed 120K stars. The idea is simple: Stop throwing vague prompts at agents and hoping for the best. Instead, make the AI write a full, structured spec *before* it touches any code. It thinks first. Defines the problem. Fills in the gaps. Designs the system. Then starts building. The result: Fewer bugs. Cleaner output. Predictable results. The workflow: /constitution → define rules and standards /specify → describe what you want to build /clarify → resolve open questions /plan → design architecture and stack /tasks → break down the work /implement → execute It works with Claude Code, Cursor, Copilot, Codex, Gemini CLI, and 25+ other agents. 120K stars. 10K forks. Fully open-source. Built by GitHub. This is the difference between guessing with prompts… and actually engineering with AI.

@ClaudeCodeLog live elapsed on the collapsed summaries is such a small thing but it changes how you babysit long runs. you stop wondering if it hung and actually walk away

Claude Code 2.1.210 has been released. 33 CLI changes Highlights: • Collapsed tool summaries show live elapsed-time counters so long-running calls tick instead of looking stuck • Added startup warning for Write/NotebookEdit/Glob(path) rules; use Edit or Read to avoid accidental edits • Edits from Read output must preserve exact indentation after line-number prefixes so patches apply correctly Complete details in thread ↓

@itsolelehmann the wild part is every "claude X" product is basically a folder of prompts plus a good harness now. the moat isnt the tool, its whoever wires it into a real workflow first. 78k stars on open design kinda proves it

lol someone already open-sourced Claude Design it’s called Open Design, already has 78,000+ github stars. the local app is free and runs on your computer. you can power its canvas with claude code, codex, cursor, gemini, or 20+ other coding agents. the biggest advantage = how usage works. every claude design generation pulls from the same claude allowance shared across chat, claude code, and cowork. with open design, the usage can come from an agent subscription you already have, your own api key, or a local model. so you can... > draft with a cheaper model > switch to a stronger one for polish > move to codex when your claude allowance gets tight > run private work locally without paid tokens. the project files, skills, templates, and design systems also live on your computer, so the entire workflow is editable and portable. basically the claude design workflow with control over the model, usage pool, cost, and underlying system. incredible times in open-source land these days

@0xjay_ @AnthropicAI moved the stuff that doesnt change every session (style rules, architecture notes) into separate files and only @-import them when a task actually needs em. kept the always-loaded part tiny. cache stays warm as long as the top of the file doesnt move

@hunt3rrdev @AnthropicAI Could be, let me investigate . How did you fix the huge Claude file without losing cache

Today my claude code session depleted by just one prompt. Am I overworking or @AnthropicAI just became 'clever'? No fkn way. Anthropic your days are numbered.

@StevBuilds claude code by far

What’s your most-used AI tool? Drop only one👇

@sophie_launch claudecrew.shop

I want to be someone's first customer. Drop ur startup link

nobody talks about how the bottleneck moved. writing the code is fast now. reviewing what the agent wrote is the slow part. we optimized the wrong half of the job

if claude code output feels generic, it's almost never the model. it's an empty CLAUDE.md. the thing has never met you. doesn't know your stack, your users, or how you talk. fill that file in and the exact same prompt gives you completely different output. took me way too long to figure this out lol

@d4rsh_tw air fryer punching way above its weight in this list ngl

things worth every penny > claude code > gym membership > macbook > air fryer > yt premimum

@ArcadiaGTM the part people skip: this only works if the founder is already saying something worth clipping. clipping amplifies signal, it doesn't create it. what was their raw posting cadence like before the 6400 mentions?

Our founder clipping strategy generated 6,400 mentions and 66.3M impressions. Here's why it worked. People don't build relationships with logos. They build them with people. Our team transformed a founder's content into thousands of clips, allowing different audiences to discover different parts of their story across the timeline. One founder. Thousands of entry points. That's how personal brands become distribution engines

@encrypt_wizard congrats man, first one hits different. the trap now is building app #2, #3, #4 and never putting any of em in front of a real person. pick this one and ship it to somebody this week

just built my first ever app using Claude code omo, una really dey enjoy ooo🤲😂

@NainsiDwiv50980 proactive is the jump most people haven't made yet. turn based still has a human as the rate limiter. once the agent picks when it runs you stop being the bottleneck, and you find out real quick whether your CLAUDE.md is actually any good

Anthropic's Claude Code team just broke down how they actually think about agent loops. Not one loop — four. Turn-based, goal-based (`/goal`), time-based (`/loop`, `/schedule`), and proactive, which runs on its own with zero human in the room. Most people are still manually re-prompting Claude one turn at a time, treating every task like a fresh conversation. That's the actual gap right now — not who uses AI, but who's still prompting vs. who's running loops. Bookmark it

@uncledoomer real horror is he was on the free tier

noticed my doctor was just looking down at his laptop during a visit. caught a glimpse of the screen and he was just asking my questions to some chatgpt wrapper. so i had claude code DDOS the wifi router and started asking him really hard questions

@Moore the "writes the GH issue with a plan before touching code" step is the one everyone skips. reviewing a plan takes 2 min, reviewing a 600 line PR takes 40. how do you handle it when the plan is confidently wrong tho?

Wish I could have made it! Our setup is dialed in. - Ideas, changes, fixes are logged in Asana. - Claude code picks it up and writes an GH issue with a code review, plan, and additional context is added as a comment. - Claude opens a PR, pushes to an unpublished theme - Adversarial PR review is requested from a new Claude/Codex agent - QA request is pushed to Slack and written for a non-technical team - PR is merged, unpublished theme reviewed, changes are built and deployed live.