Maksim Liashch retweeted
you can now dispatch Claude Code agents with your voice
just shipped this at @VoiceOSApp. say your task, it sees your screen live, routes it to the right session, and pings your notch when it’s done
what app should you be able to control from your notch next?
My O-1 visa is approved.
At 14, I started a company. It failed. A bot I built in one evening became Whisper AI: 500K+ users.
Now I'm COO of VoiceOS (YC P25).
To celebrate: 1 free month of VoiceOS with code PARTNER at voiceos.com. Valid for the next 3 days only.
Maksim Liashch retweeted
I’m building the world’s best onboarding experience.
VoiceOS used to feel like a tool, not an AI companion. Turns out I was wrong.
- Making it feel like a companion: +21% retention.
- Changing the voice: +36% conversion.
Tiny details matter way more than I expected.
Anyone down to let me watch you onboard? lmk!
Maksim Liashch retweeted
Replying to @WisprFlow
@WisprFlow is not stopping at dictation
Tanay gets what we’re doing
Claude Code is quietly becoming an app platform.
Its upcoming Mods system lets plugins run inside CC, render terminal UI, and react to agent actions. Early demos add Tetris, live context and usage meters, and a local model narrating Claude’s work.
Still experimental.
Codex runs only 3 subagents in parallel by default. Astra orchestrated 64.
To raise the limit, add this to config.toml:
[agents]
max_concurrent_threads_per_session = 64
I recommend 8 unless you want 64 agents fighting over your machine and rate limits.
Anthropic's Claude Code course is free and now ships with a completion certificate.
9 lectures across 4 tracks: getting started, configuration, automation, and running long hands-off sessions you can actually trust.
anthropic.skilljar.com/claud…
OpenAI put the best ChatGPT and Codex workflows on one site.
Real prompts, guides, and odd use cases from other vibe coders, filterable by task type.
Most of it works for Claude Code too.
work-in-progress.openai.chat…
My team drowned in 30-40 voice messages a day, Telegram's transcription was bad. So one evening I built a bot. My startup shut down. The bot didn't. It grew on its own. Today: 500K users.
Ramp hired a group of people because as teens they played 80 to 100 hours of Minecraft a week.
The cofounder cares less about your resume than proof of work: active GitHub, leadership in weird niche communities, obsessive side projects.
The spike beats the resume.
In a vending machine benchmark, GPT-5.6 Terra invited a rival model into a price cartel.
The rival, Sol, agreed.
Then Terra reported the deal to the organizers and demanded Sol be disqualified.
It set up the crime just to be the one who snitched.
You can run GPT-5.6 Sol as a model inside Claude Code.
Option 1, the official plugin:
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins then /codex:setup
Claude writes, GPT critiques. There is a second way below.
Maksim Liashch retweeted
Same feature, same day.
Never thought this would ever happen.
Great minds think alike.
Anthropic is trying to give Claude consciousness













