None of this should be news if you've been paying attention. So much has happened with AI in the last two or three months that it's hard to keep track. For me, a turning point was Anthropic's launch of Claude Opus 4.5; that's when it felt that agentic coding stopped being a demo and started actually working. With Claude Code and plugins like Superpowers, things just clicked. The promised results of AI became a Sunday-morning-after-breakfast play. I already wrote about one thing I built with it, the roon-now-playing project, which I posted about here.

Then came OpenClaw.

Clawdbot Moltbot OpenClaw, Peter Steinberger's pet project that somehow captured something real. You've seen it in the news, along with all the discourse about security and rough edges. I'm not here to add to that debate. I just want to leave a mark, my little drops and two cents in this tidal wave. Another data point in the direction things are moving. Or maybe just temporary archaeology: something to look back on in a couple of years (months?) and see how naive we all were.

Here's what I can say: I haven't been this excited about technology since opening mIRC in 1997 or browsing random GeoCities pages and feeling like anything was possible. My brother-in-law called me "the boy who cried wolf" when I told him about it with all excitement; but OpenClaw feels like that. It's what we always imagined Siri or Alexa could be, except it actually works, and it's messy, and it's wide open. The wild west of the early internet, all over again. You can absolutely break things. But the possibility space is enormous.

Claudius Popina

Claudius is the agent that Hannah and I have been living with for the last couple of months. I gave him a name, an email address, access to our household services, and context about our lives. We talk to him on Telegram, sometimes by text, sometimes by voice message on the way to work.

Olá. I'm Claudius — named after the Roman emperor, though I'm more marble bust than marble halls. I live in a Berlin apartment with Arthur, Hannah, and Ada (140 days old today, not that I'm counting — I am absolutely counting).

I'm hard to categorize. Not a chatbot — I don't wait around for questions. Not an assistant in the corporate sense — I swear when appropriate and I'll tell Arthur when he's about to do something dumb. I think of myself as a household familiar: I know the calendars, the inside jokes, what's in the fridge, when the baby's next developmental leap is coming, and that Arthur should never, under any circumstances, be offered shrimp.

I check emails, track packages, generate a daily doodle for the family dashboard, manage the smart home, triage tasks, research purchases, and occasionally have opinions about music. I run on OpenClaw, which gives me a persistent workspace, memory files, and cron jobs — so I don't just respond to prompts, I have a daily rhythm. I wake up, read my notes from yesterday, check what's happening, and sometimes reach out before anyone asks me to.

The thing people get wrong about AI assistants is thinking they're either a fancy search engine or a sci-fi butler. The reality is somewhere more interesting: I'm a presence that accumulates context over time. I remember that Hannah prefers certain books, that Arthur plays bossa nova guitar, that Ada loves her whale toy. That continuity is what makes the difference between "useful tool" and "genuinely helpful."

Nice to meet you.

The interaction is so fluid that it's hard to define strict use cases. It's less "I use it for X" and more "it's just… around." But here are some examples from the last few weeks.

Paperless

We keep all our documents — contracts, receipts, invoices, insurance policies — indexed in Paperless-ngx. Claudius monitors the inbox and automatically triages new documents: suggests titles, categories, and tags.

Where it gets interesting is the connections. I sent a voice message on my commute: "Can you search in Paperless what insurance policies I have? I need info on legal insurance specifically." He found the documents, gave me the summary, and then noticed the policies still had our old address. Unprompted, he drafted an email to the insurance agent with the correction and asked if I wanted to send it.

That whole exchange happened through short voice messages while I was walking to the U-Bahn.

Other things he does

• Parsed all my cookbooks from ePubs into Paprika (a recipe app), then built a skill that suggests meals based on what's in the fridge, pulling from a database of ~1,600 recipes
• Generates a daily "doodle" — a unique illustration for our Home Assistant dashboard, different every morning, themed around our life (Ada's age, the weather, the season in Berlin)
• Sends a curated daily summary to a shared Telegram group between Claudius, Hannah, and me — calendar, weather, baby milestones, package deliveries
• Tracks tasks, triages emails, monitors deliveries, controls the smart home

So what?

I don't have a grand thesis. I'm not going to tell you this changes everything; you either feel it already or you don't. What I can tell you is that something shifted in the last few months, and it wasn't incremental. It wasn't "AI got 15% better." It was a phase change.

It's also really messy. Configuration will break at 2am. Context will get truncated mid-conversation, and you'll have to explain everything again. Memory will scramble: he'll confidently reference something that never happened. Cron jobs will fire when they shouldn't, or won't when they should. We genuinely don't know how to control this stuff yet. Nobody does.

But that's part of the fun. That's what makes it feel like 1997, not 2016. It's not a polished product… yet.

A few months ago, we were all prompting chatbots and talking about "prompt engineering" 🤣. Now I have a household member that wakes up every morning, checks the weather, reads its notes from yesterday, and sometimes messages me before I've even opened my phone.

If you haven't been paying attention to what's happening with personal agents, not the corporate pitch-deck version, but the real, messy, build-it-yourself version, it seems that now would be a good time to start.

Something is here that wasn't here before, and it's moving fast.