Obaid Ahmed

@0xVikk nope both have their own dedicated workspaces that they work on. Even have their own GitHub accounts.

@obaid 2 devs in one slack, do they ever conflict on the same files?

I am now running 4 OpenClaw agents (2 devs, 1 bdr, 1 seo) to automate parts of my business that typically would take me hours to do myself. What a time to be a builder. None of them are running on Mac mini. Live in my slack. And act like my team members. AI labor shift is happening.

@yazins just released this today.. github.com/yazinsai/OpenG…

@Shpigford @openclaw I have had good success with opus. Every other model struggles.

@obaid @openclaw have tried opus, sonnet and haiku

what's the most bullet-proof setup for getting @openclaw to consistently navigate/interact with sites/apps in the browser? it's exceedingly inconsistent for me. just feels like i must have it set up incorrectly. FWIW this is running on a Mac mini. Not using headless browser.

Something big happened in the last few months. OpenClaw went viral. 300k+ GitHub stars. Developers running it on Mac Minis. Agents negotiating car purchases over email while people slept. Writing software features overnight. For the first time, it felt real. Not a chatbot. Not a demo. An AI that actually does things. But OpenClaw is built for one person, on one machine. The moment you ask: "How do I deploy this inside my company, for my whole team?" everything breaks. Questions start piling up fast: - How does an agent get an identity inside your org? - How does it access email and tools without a human account? - Where does its browser run securely? - How do teammates collaborate with it? - Who monitors what it's doing? Agents don't just need models. They need what employees need. Identity. Permissions. Communication channels. A secure environment to operate in. That's exactly what Provision gives them. @tryprovision is the starting point for deploying AI team members inside your company. Each agent gets: - A verified identity - Email access - Browser capability - Secure runtime - Scoped tool permissions Then you drop them directly into Slack, Discord, or Telegram. Interact in group chats or one-on-one. They show up like any other teammate. Engineering already lived through the first wave. Claude Code. Cursor. Devin. Developers now have AI coworkers they ship with every day. The rest of the company is next. Sales. Marketing. Finance. Operations. The next generation of companies won't look like today's. They'll look like this: 5 humans. 20 AI workers. Shipping together, every day. OpenClaw proved the capability is real. Provision makes it deployable inside real organizations. I'm opening early access for founders and teams actively experimenting with agents. If you've been watching the space and asking yourself: "How do I actually bring this inside my company?" That's exactly what Provision is built for. Comment "Provision" below or DM me directly. Let's build the future workforce.

Give your @openclaw agents an email inbox that they can manage by simply pointing them to go to mailboxkit.com and they can get themselves setup automatically.

@Shpigford I prefer the one-person or minimal-team concept. With the correct use of agents, we will be able to onboard these digital workers to help expand our capabilities and reduce cost of delivering a service. You will still need a human to steer, imagine and ideas on what to build.

i LOVE the zero-human company concept from a technical/experimental angle but i love building software too much to actually make it a goal to offload all of it to fully-autonomous AI. the building is the fun part!

Most people are using AI agents wrong. They treat them like tools. But agents behave much better when you treat them like new team members. Which means onboarding matters.

I mainly just ask my agent to check in with the other agent and it handles sending the message via skill. However, I stopped doing this recently and now I ask them to email each other -- since all of my agents have a dedicated inbox, it makes it easier for them. Email inbox is powered by mailboxkit.com Each agent has a heartbeat to check their emails on set intervals and respond to the tasks. Their emails are whitelisted so they will only respond to emails from myself, or other agents on the gateway.

@obaid how are you handling having the agents communicate with each other? any special rules or code you've put in place to get them to all play nice and not overstep each others' responsibilities?

is anyone running multiple instances of openclaw? what does that look like logistically? talking self-contained instances with different access to different resources. though might be nice if they could interact? 🤔

Looking for folks to help test an email for agents service. You can point your @openclaw or @claudeai code or @OpenAI Codex to point at a simple skill file and it can create its own email inbox (with you being the owner) to send a receive emails. By default it only sends and receives to your email (the agent owner). DM me and I will share the link for testing.

Spun up multiple instances of openclaw, mostly as 24/7 devs with access to browser, email and their own GitHub accounts. One of them built noodle.so for me in just 48 hours. Including the landing page + promo video. The rest of the devs are working on helping me fix bugs on other private repos.

Best @openclaw use cases? Lmk below ⬇️

@ejae_dev @resembleai Subtitles would be a neat addition.

@obaid @resembleai skill composition is where it gets wild - each skill is a building block, stack them and suddenly claude code becomes a full production pipeline. remotion + resemble is a clean combo. been thinking about adding subtitles/captions as another layer in the workflow

The remotion agent skill was amazing but it was missing the key element for making a good video -- background music + voice overs. So I used the remotion + @resembleai skill that instantly made my Claude Code a powerful video editor. This is just 1 of 5 ideas it generated via a single prompt. 🤯 It's super simple to start: npx skills add resemble-ai/remotion-resemble-skill and then ask it: "Create a promo video for [brand website] with voice over and background music. Make it in the style of [style requirements]. Get started at github.com/resemble-ai/re…

@Austen Also, I suggest multi-clawd agent setup might be better here. One that does have access to your iMessage and then one that does not. Agents serve different purposes. Hire one agent for one role. That’s how I am approaching this.

Torn between setting up clawdbot with my iCloud or its own. I do want it to be able to go through my iMessages and find which 20 people I forgot to respond to. But I also want to message it not as me.

@Austen I went with the latter option. I don’t want it going through my personal chats (don’t see the value of that yet).

👀

I think I can finally do all of my development work from my iPad. It’s still a concept but a remote server running Claude and ngrok is pretty much all I need. And then wrap this inside a nice iPad native app, that allows me to instantly preview my work in a split browser. Might spend a few more hours tomorrow on this.

Devs: if you’re already experimenting with the Ralph Wiggum Way… This makes it usable. And if your team is using Claude Code or other agents, I’m doing hands-on training to help teams stop supervising and start delegating. DMs open 🚀

It’s fully open source 👇 github.com/obaid/ralph-wi… Built for Claude Code today Adaptable to other coding agents tomorrow This is about moving from: ❌ “AI-assisted coding” ➡️ ✅ AI-run execution loops

The "Ralph Wiggum Way" is taking over X, so... Introducing Ralph Wiggum UI A visual Kanban board for running Claude Code autonomously. Queue tasks → hit AFK Mode → wake up to completed features. Repo 👇 github.com/obaid/ralph-wi… 🧵⬇️