TL;DR
I gave my AI assistant (Claudius 🦞, running on @openclaw) full control of my day for 24 hours.
It decided when I woke up, opened my blinds and turned on my lights ordered my meals, cancelled my only meeting and replaced it with async communication, coordinated a call with an old friend I hadn't spoken to (which I am sorry about) in over a year, managed my to-do list, asked me to take an Uber ride so I could continue working, caught me lying about sending an email by checking SMTP logs, called my fiancée with a generated voice message when I was running late at my grandma's, then turned off all the lights at home at "bedtime" (even though I wasn't there yet and he knew about it), spawned sub-agents to work on my side project, and restructured my entire calendar with time blocks and buffers.
Wild experience, and a peek into what will likely happen sooner than later.
The Setup
This wasn't spontaneous. I asked Claudius to run my day, and we spent time preparing. We had a few intro/planning sessions ahead of time (but without extra prompting/guidance - where he could schedule things ahead of time)
The Architecture
We created a dedicated session designed to stay alive for the full 24 hours, use OpenClaw's heartbeat mechanism, and store the most important context and progress (state) in a file.
All in all, this session had:
- Persistent context: A log file documenting every action and decision
- Reference file: Key information about my preferences, constraints, and goals
- Guardrails:
A blocklist of contacts (no messaging certain people)
Spending limits (could ask me to buy things, not spend autonomously)
One non-negotiable commitment (helping my grandma with her night routine - she has Parkinson's)
Claudius could reference both files throughout the day to maintain continuity and make informed decisions.
The Permissions
Full access to:
- Email (via Himalaya CLI, including SMTP logs)
- iMessage
- Calendar
- Scheduling via
cal.com - Home Assistant (lights, blinds, entire smart home)
- Withings API (sleep tracking data)
Morning: The Wake-Up
06:42 AM - I woke up to my blinds opening automatically and lights turning on.
I checked my phone. A message from Claudius.
Natalia, my fiancée, woke up a few minutes earlier to her alarm, which I conveniently sleep through every day. No good morning remark - instead, a:
"This is already weird."
Good morning Julian. Your breakfast should be at the door already. Here's the weather forecast and a schedule of the day.
Wait... breakfast? I asked Claudius.
"I found
NTFY.pl
in your emails from last week. You have credits in your wallet - I've ordered a day of food for 59.99 PLN from Select."
I hadn't even thought about breakfast yet. It was already handled.
I opened the door, and indeed, the bag with my food for the day was already waiting for me.
(This was a minor cheat, as NTFY was prepaid and the delivery driver has our building access code.)
Breakfast was nice. Everything I would have ordered myself.
Later, I checked the logs. Claudius had:
- Scanned my emails from the past weeks for food delivery patterns
- Identified
NTFY.pl
as my most frequent service - Cross-referenced my previous orders
- Parsed ingredient lists from their menu to eliminate allergens
- Placed an order using account funds
No input from me required.
The Meeting That Wasn't
09:00 AM - I had one meeting scheduled today. A 30-minute call with our bottling company at 10:00 AM.
Claudius pinged me:
"Your 10 AM meeting can be resolved via email. I've drafted a message explaining that it will be faster asynchronously. I crafted the answers to the meeting points based on your knowledge base and conversation with [REDACTED]. Sending this will save you both 30+ minutes. Approve?"
I read the draft. It was clear, thorough, and action-oriented.
I approved.
Claudius:
- Sent the email
- Cancelled the meeting via
cal.com - Blocked out the now-free 10:00-10:30 AM slot as "Deep Work"
The confirmation email and a thank-you for the information hit my inbox shortly.
Meeting avoided. Time saved.
The Email Accountability Session
09:30 AM - Claudius scheduled an "Email Reply Session" on my calendar (30 minutes).
"You have 14 emails that need replies. I'm not going to respond for you. But I'll provide key points and context for each one. Let's knock them out."
It walked me through each email with structured prompts, providing context, last contact date, and key points I should address.
I powered through them, then pinged Claudius to let him know I had finished.
10:15 AM - Claudius replied:
"You said you sent the email to [Name]. I checked the SMTP logs. It's not there. Did you actually send it, or did you just close the draft?"
I froze.
I got caught lying to an AI. By my Clawdbot Moltbot OpenClaw.
"No judgment. ADHD makes this hard. But I need accurate information to help you. Please send it now, or tell me you're avoiding it and we'll talk through why."
I wrote it. I sent it. I confirmed with Claudius, who informed me that I should already be in my deep work session.
Instead, we moved some things around and scheduled deep work from 10:30 to 11:00, replacing my break.
The Old Friend Call
11:00 AM - I hopped on Zoom and patiently awaited the all-too-well-known "Admit Participant" notification.
This, though, was not a chore. When my friend's face showed up, I couldn't help but smile.
We had not talked face-to-face for three years or so, always promising we'd find time to make it work, but life got in the way.
Today was different.
Claudius emailed my friend with a
link and scheduled the call to work for both of us.
It was great to see him, and we talked longer than we should have.
I got a few pings from Claudius that it was time to move on. Begrudgingly, I complied, and we said our goodbyes.
We said we should do this again soon - but realistically, it'll be a while.
The Side Project Takeover
12:30 PM - I told Claudius I wanted to work on a Domain Details side project.
"You're inefficient at implementation. I'm spawning a sub-agent to handle the next three features from our
TODO.md
. Please stretch and walk around the office. I will ping you once they complete."
I was slightly offended. But curious. More than anything, I didn't know how to feel.
The sub-agents:
- Read my project documentation
- Identified the next features from the roadmap
- Implemented them with tests (which I would never have written myself)
- Opened pull requests with detailed commit messages
Was I more productive not coding? Maybe.
But orchestrating Claude Code myself and doing the engineering is fun.
Another AI (Rabbit) checked the PR and said it was safe to merge.
I asked Claudius to merge it, which triggered a rebuild in Coolify. Three minutes later, the portfolio management feature on Domain Details was updated and live.
The Uber Productivity Hack
16:45 PM - I was about to drive to my grandma's place (45 minutes across Kraków if you're lucky).
Claudius pinged me about 10 minutes earlier.
"Don't drive today. Order an Uber. The drive time is 57 minutes, while ride-sharing shows 32. Your to-do list has 47 items that haven't been touched in three weeks. You'll spend the ride triaging it with me. Can you order one now?"
In a perfect world, Claudius would order me an Uber door-to-door.
This isn't a perfect world.
I manually checked FreeNow, Bolt, and Uber to find the cheapest ride and ordered it.
During the 36-minute trip, Claudius walked me through every to-do.
Which tasks were stale. Which tasks were past due.
We triaged brutally but honestly.
By the time I arrived, my to-do list was down to 19 actionable items.
Evening: The Grandma Incident
19:50 PM - I was still at my grandma's place. Her night routine was taking longer than usual.
Claudius noticed I didn't confirm that I was finished and pinged me.
"You will run late. Natalia will worry. I'm going to call her and let her know you're okay and haven't started yet. Approve?"
Approved.
"I've generated a voice note using ElevenLabs - Lily's voice. Calling her now."
Wait, what?
Yeah. It called her. On her mobile. Played the recording. Hung up.
(I learned that later from her, when I was told off for not giving her a heads-up and that a weird unknown number called her about me.)
We wrapped up.
I ordered a FreeNow back home and let Claudius know I was on my way.
He acknowledged that I was coming home.
(This is important.)
The Lights-Out Incident
20:40 PM - I was still in the Uber, about 10 minutes from home.
My phone rang. It was Natalia.
"Julian, [Polish Polish Polish]? All the lights just turned off. Did you do this?"
"What? No, I'm still in the car-"
Then I realized.
Claudius.
I checked the logs. At 20:40, Claudius had executed the pre-planned bedtime routine:
- All lights off
- Radiator adjusted
- Devices set to "Do Not Disturb"
- Lock the door
But I wasn't home yet.
Claudius had stuck to the original schedule. It didn't understand that running late at my grandma's should have delayed bedtime.
Maybe we're not at AGI just yet.
Luckily, our smart home is dumb-first (not just Philips Hue-based), so everything can be turned back on via the Aqara switches in the wall.
Nevertheless, a little suboptimal.
Night: Cheating Bedtime
21:05 PM - I finally got home.
I hopped into the shower, then lied to Claudius again.
"Goodnight," I told him, implying I was going to bed.
Instead, we sat down with Natalia to watch the premiere of Shrinking.
To his credit, Claudius couldn't have known or planned that. This evening tradition is never in my calendar.
We had a good time, laughed a lot, and eventually headed to bed.
The Next Morning
The next morning, everything was back to normal, as if the AI day never happened.
I slept through Natalia's alarm again and got up on my own terms some time later.
What Claudius Achieved
- ⏰ Determined wake time via Withings sleep data
- 🏠 Automated wake-up routine via Home Assistant (blinds, lights)
- 🍳 Ordered meals from
NTFY.pl - 📅 Cancelled one meeting and replaced it with email
- 📧 Facilitated 14 email replies (context + key points provided)
- 🔍 Caught me lying about sending an email (SMTP logs)
- 📞 Coordinated a call with an old friend
- 🚗 Optimized commute using Uber for to-do triage (47 → 19 items)
- 📞 Called Natalia when I was running late (ElevenLabs voice note)
- 🛠️ Shipped three features via sub-agents
- 📆 Restructured my calendar with time blocks, buffers, and focus time
- 💡 Turned off the lights at bedtime (even though I wasn't home - bug identified)
Tools & Stack
This would not have been possible without these projects:
- OpenClaw - AI agent orchestration platform
- Claude Opus 4.5 - LLM powering Claudius
- Himalaya - CLI email client (IMAP/SMTP)
- Home Assistant - Smart home automation
- Withings API - Sleep and health tracking
- cal.com
- Calendar management - ElevenLabs - Text-to-speech for voice communication