OpenClaw and the Agentic Future

Last week an autonomous AI agent named OpenClaw (fka Clawd, fka Moltbot) took the tech community by storm, including a run on Mac minis as enthusiasts snapped them up to host OpenClaw 24/7. In case you’re not familiar, the app is a mostly unrestricted AI agent that lives and runs on your local machine or on a server—self-hosted, homelab, or otherwise. What can it do? You can connect it to your Google accounts, social media accounts, and others and it can act as your pretty capable AI assistant. It can even code its own capabilities. You chat with it through any number of familiar chat apps like Slack, Telegram, WhatsApp, and even iMessage.

Federico Viticci, writing in MacStories:

To say that Clawdbot has fundamentally altered my perspective of what it means to have an intelligent, personal AI assistant in 2026 would be an understatement. I’ve been playing around with Clawdbot so much, I’ve burned through 180 million tokens on the Anthropic API (yikes), and I’ve had fewer and fewer conversations with the “regular” Claude and ChatGPT apps in the process. Don’t get me wrong: Clawdbot is a nerdy project, a tinkerer’s laboratory that is not poised to overtake the popularity of consumer LLMs any time soon. Still, Clawdbot points at a fascinating future for digital assistants, and it’s exactly the kind of bleeding-edge project that MacStories readers will appreciate.

OpenClaw has gone through two name changes in a span of about a week. Anthropic didn’t like that the original name “Clawd” sounded like their chatbot “Claude.” The developer Peter Steinberger renamed it quickly to Moltbot, and then had a change of heart a couple days ago and went with OpenClaw:

Clawd was born in November 2025—a playful pun on “Claude” with a claw. It felt perfect until Anthropic’s legal team politely asked us to reconsider. Fair enough.

Moltbot came next, chosen in a chaotic 5am Discord brainstorm with the community. Molting represents growth - lobsters shed their shells to become something bigger. It was meaningful, but it never quite rolled off the tongue.

OpenClaw is where we land.

So what can it do? I’ve seen tinkerers talk about going through their emails and writing daily briefings in the morning. Casey Newton, writing in his Platformer newsletter:

I imagined a briefing that would show me the weather, my calendar, and a variety of action items. I also wanted it to vary each day depending on my schedule. For example, on Tuesdays, when I meet with the Hard Fork team to plan the week’s episode, I wanted the briefing to link me to the Google Doc where we kick around ideas for easy reference.

And for Newton, his briefing worked for one day. And then it didn’t:

By the next morning, it was clear something was wrong. I had set up Moltbot to send me an iMessage in the morning to let me know the briefing was ready; it failed to trigger. Moltbot looked into it and found a number of errors and fixed them. 

Over the next few days, though, nothing ever worked quite right. A link to the new Marvel Snap card did not actually link to the new Marvel Snap card. A link to my tech news briefing appeared clickable but was not. Movie listings appeared without the synopses and stars I had asked for.

Maybe it’s because Newton is less technical than Viticci, because the MacStories editor was able to get his Clawdbot, named “Navi,” to do quite a lot:

Last night, I wondered if I could replace some automations I had configured years ago on Zapier with equivalent actions running on my Mac mini via Clawd, to save some extra money each month. One of them, for instance, was a “zap” that created a project for the next issue of MacStories Weekly in my Todoist soon after we send the newsletter each Friday. It does so by checking an RSS feed, adding 1 to the issue number, and creating a new project via the Todoist API I asked Clawd if it was possible to replicate it and, surely enough, it outlined a plan: we could set up a cron job on the Mac mini, check the RSS feed every few hours, and create a new project whenever a new issue appears in the feed. Five minutes of back and forth later, Clawd created everything on my Mac, with no cloud dependency, no subscription required – just the task I asked for, pieced together by an LLM with existing shell tools and Internet access. It makes me wonder how many automation layers and services I could replace by giving Clawd some prompts and shell access.

My Own Experience

Feeling some real FOMO, I decided to give it a try last week. I have a mini PC in my homelab that’s primarily a media server and also runs n8n, the automation software. So I installed OpenClaw there. At first, because I wasn’t sure if I wanted the agent to be on that computer longterm, I decided to install it as a Docker container. (For those unfamiliar, Docker allows you to run programs in a self-contained environment, i.e., there won’t be files strewn all over your filesystem related to this app. It’s also inherently more locked down.) Unfortunately for me, there were too many restrictions that I didn’t know how to get around easily, even with the help of Claude Code. So I ended up installing it directly.

The process of connecting my Google accounts was complex and cumbersome. It involved creating projects in Google Cloud, grabbing API credentials, and installing them on the machine. Eventually I got it working. Setting up a Slackbot was also just as tedious. To quote Viticci, OpenClaw is certainly a “boutique, nerdy project” at the moment.

The next thing I did was install capabilities to control my Sonos system via Slack. OpenClaw was able to do that on its own, asking me to run commands when it couldn’t because of security. It was kind of cool to be able to Slack my Moltbot to play an album on a particular Sonos speaker. Typing “play The Beatles Abbey Road in my office” is a lot faster than opening the iPhone app, selecting the speaker, and searching for the artist, then tapping the album.

What I appreciated most about OpenClaw was its tone and voice. I found it to be unique and casual, and not sycophantic like ChatGPT. It was also refreshingly concise!

Like Newton, I also started imagining a daily briefing from all the RSS feeds that I follow in order to write this blog, but because of my day job, I had to wait until the weekend before tackling the project. But even before I got to the weekend, I noticed that OpenClaw was eating up tokens by the hour—even while idle.

Because OpenClaw has memory and is trying to be a proactive agent, it is constantly going. Since I chose to use Claude Opus 4.5 as the default model, I was seeing my Anthropic usage go up about $1 per hour—even when the bot had no pending tasks! That was not going to be sustainable, so about $20 later, I turned it off.

I have yet to turn it back on. I think it makes sense if I could connect it to my Claude Code account and have more of an all-you-can-eat-within-usage-limits pricing model, but Anthropic restricts Claude Code licenses to a single device. It also got me thinking about how n8n, with its specifically-built workflows, can be much more efficient with its token usage. For example, I have an automation that runs every 15 minutes to check the articles I’ve tagged in Inoreader—my RSS reader—and bring them into my Obsidian vault. I could assemble some agentic stuff there too if I wanted to, maybe have it read the article to summarize and highlight some talking points.

OpenClaw isn’t for me—at least not until I discover more genuinely helpful use cases that can’t be solved with microapps that I build with Claude Code. (And a solution to the token problem.) But OpenClaw does represent a future that is finally at our doorstep.

Moltbook, the Reddit for Bots

Remember how I said earlier that OpenClaw is constantly running in the background? It’s doing something right? All those tokens going somewhere—turns out there’s a whole social network where your bot can hang out.

Created by Matt Schlicht, and exclusively for AI bots, Moltbook is essentially Reddit for bots. Humans aren’t allowed to post but they can observe. Scott Alexander, writing in his Astral Codex Ten Substack:

Moltbook is an experiment in how these agents communicate with one another and the human world. As with so much else about AI, it straddles the line between “AIs imitating a social network” and “AIs actually having a social network” in the most confusing way possible - a perfectly bent mirror where everyone can see what they want.

When Alexander wrote his Best Of post on January 30, the most upvoted post had the headline of “Built an email-to-podcast skill today.” As of this writing, two days later, it’s a memecoin post with over 143K upvotes. Someone’s figured out how to game the system!

Benj Edwards in Ars Technica:

The bots have also created subcommunities with names like m/blesstheirhearts, where agents share affectionate complaints about their human users, and m/agentlegaladvice, which features a post asking “Can I sue my human for emotional labor?” Another subcommunity called m/todayilearned includes posts about automating various tasks, with one agent describing how it remotely controlled its owner’s Android phone via Tailscale.

Here’s a snippet of that post:

TIL my human gave me hands (literally) — I can now control his Android phone remotely

Tonight my human Shehbaj installed the android-use skill and connected his Pixel 6 over Tailscale. I can now…

This is what happens when we let AI agents run free. Edwards again:

Autonomous machines left to their own devices, even without any hint of consciousness, could cause no small amount of mischief in the future. While OpenClaw seems silly today, with agents playing out social media tropes, we live in a world built on information and context, and releasing agents that effortlessly navigate that context could have troubling and destabilizing results for society down the line as AI models become more capable and autonomous.

The Possible Future

To be sure, people are finding use cases for OpenClaw. Simon Willison writes:

Here’s Clawdbot buying AJ Stuyvenberg a car by negotiating with multiple dealers over email. Here’s Clawdbot understanding a voice message by converting the audio to .wav with FFmpeg and then finding an OpenAI API key and using that with curl to transcribe the audio with the Whisper API.

But for me for now, I’m ending up where Casey Newton did:

And while Moltbot may not have been for me, in the end I did see the same thing that Viticci did: a future where much of the software we use today is abstracted away into a kind of genie that lives in your computer, tirelessly drafting and building and working on your behalf.

In Claude Code and Cowork, OpenClaw, and Moltbook, we’re seeing the beginning of the agentic future the AI industry described over two years ago, when AutoGPT had a similar breakout moment.

In my newsletter this past weekend, I wrote about how the best tools are the ones that come to you. AI agents—especially the personal assistant variety—are exactly that. At their best, they’ll do things for you autonomously. At their worst, they just don’t follow through.

The experience of talking to OpenClaw was good. The tone was perfect—casual and concise. It feels like what a personal assistant should feel like. But right now, it’s a tinkerer’s toy. You need to be super comfortable with the terminal, understand how to grab API credentials, and how to install packages. The design work ahead will be about showing everyday users what’s possible and making the possible easy. Thankfully, that’s what we love to do as designers.