AI Systems Dev

🔁 Resumable Sessions One of our most-requested features is here. You can now resume past sessions and continue working exactly where you left off without restarting or re-explaining. Just click a past session and keep going with all context intact. Let us know how this

@steipete What is this 😎

@sequoia @steipete @openclaw @LucianaLix Inspired

At our @OpenClaw meetup, @steipete talked with @LucianaLix about the question that led to the creation of OpenClaw – the same question behind the founding of many now-legendary companies: “How hard can it be?”

@nummanali Nice

OpenAI has acquired uv, ruff and ty

@PawelHuryn Nice

My Claude hype has been doubling every day for a week now 😅

@steipete @openclaw 🔥

new @openclaw beta is up! gollum in the VM approves. github.com/openclaw/openc… What's new? "stop openclaw!", Android refresh (yes we have apps for iOS Android macOS Windows and all, just not quite ready for prime time yet), Big reliability fixes for cross-channel routing, Heartbeat is safer by default (no more DM leaking), Discord reliability improvements, WhatsApp safety/reliability, some macOS work and security hardening (lots of allowList tweaks this time)

@PawelHuryn Interesting

Everyone's building "agent swarms." Nobody's talking about why they work. It's not the orchestration. It's the context allocation. A solo dev just documented his setup: one AI holds business context (customer data, meeting notes, past decisions). Separate AIs hold the codebase. The orchestrator writes prompts. The workers write code. Context windows are zero-sum. Fill it with code, no room for customer context. Fill it with customer history, no room for the codebase. So you split the tiers. The orchestrator understands customers, priorities, context. It writes the prompts. The coding agents understand the repo, conventions, tests. They write the code. That's not a PM and an engineering team. That's a backlog manager and a squad of coders who follow instructions. But the orchestrator doesn't decide what's worth building. It translates decisions that already got made. The coding agents don't question the spec. They ship it. Agent swarms automated the bottom two tiers of building software. Task scoping and code generation. What remains is what was always hardest: figuring out which customer problem matters most and what to say no to. Agents didn't touch product management.

@steipete @openclaw 👀

New CHUNKY @openclaw beta is up. Will wait a few h before flipping the switch to catch regressions. Reply to this tweet if you find any blockers that were not in .21 github.com/openclaw/openc…

@nummanali Agree X is high signal for agentic work. How do you account for selection bias, builders who learn elsewhere or don’t post publicly?

Going to sound controversial here But if your next candidate is not active on X Then they are probably no good for you I’ve found those informed through X perform significantly better with agentic engineering than those not You simply can’t find the info here elsewhere

I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞 steipete.me/posts/2026/ope…

@steipete 😎

New version of summarize.sh is out! 0.11 can now even use cursor for free tokens + uses Groq for way faster TTS inference + lots lots of other improvements. github.com/steipete/summa… summarize youtube.com/watch\?v\=n1E9IZfvGMA --slides my favorite way to consume YT. Or Podcasts. or literally any website or remote or local file. Powers the content efficient file summarization for @openclaw but also really nice for humans, especially with --slides. It's also a Chrome extension. because why not. chromewebstore.google.com/detail/summari…

RT @PawelHuryn: OpenClaw has 186K GitHub stars and 1.5M compromised API keys. I needed a secure alternative. So, I built it with n8n and C…

@PawelHuryn Awesome

We successfully trained an LLM without human intervention using Claude Code. We made a guide on how to do this with local LLMs via Claude Code and OpenAI Codex. Connect GLM-4.7-Flash to your server and start agentic coding locally! Guide: unsloth.ai/docs/basics/cl…

Tell all your friends All your family If they're not using AI Agents Now is the time Get them to use GPT 5.3 Codex if they're on ChatGPT Get them to use Opus 4.6 if they're on Claude Do the world a favour and spread the word Don't let them be left behind

@nummanali 💯

@nummanali 🔥👏🏼

Omg it is GPT 5.3 Codex How do we get that before GPT 5.3!? We are so spoilt today - Opus 4.6 and GPT 5.3 Codex World of software going to melt down today

@tinkerapi @miramurati 👊🏼

Since Tinker launched, our community has used it to train state-of-the-art models, build infrastructure, and publish novel research. We will be highlighting this creative work in regular roundups, and hope to inspire your own Tinkering as well.

@OpenAIDevs @romainhuet 🔥👊🏼