Klemens Arro

iMessage conversation with Andy. Andy: “Looks like food wasn't logged today — hard to track the deficit without it. Even a rough log helps.” (2m) Reply: “It was. But I guess the sync hadn’t run yet. Is the data there now?” (Now) Andy: “Could you remind me what you're referring to? I want to make sure I check the right data — the last few topics were moving quotes, health stats, and earlier the morning brief. Which sync are you asking about?” (Now)

These agents' confusion moments are a great example of this other 20-30%.

👉 https://ailab.ee/we-turned-openclaw-into-a-fully-functioning-executive-assistant/

Banner message: “Your access token could not be refreshed because your refresh token was already used. Please log out and sign in again.” Below, a status panel titled “We’re currently experiencing issues” (ChatGPT) listing: “Increased errors on ChatGPT File Uploads” — “Another issue has been identified and a mitigation is being applied.” “Investigating · Ongoing for 18 hours · Affects ChatGPT”; “Increased errors with ChatGPT file downloads” — “We are still implementing the mitigation.” “Identified · Ongoing for 16 hours”; “Codex unresponsive” — “We have applied the mitigation and are monitoring the recovery.” “Monitoring · Ongoing for 2 days · Affects ChatGPT.”

Haha. We need more strong players in the coding field. I figured that I’d use Codex then, while Claude is down... well, no luck here either 😂 I’m afraid to open Antigravity now.

Terminal screenshot showing authentication failures and a status update. Error text includes: “API Error: 401 {“type”:“error”, “error”:{“type”:“authentication_error”, “message”:“Invalid authentication credentials”}, “request_id”:“req_011CYwXGnJEv6qaFasxYwi2S”} · Please run /login” and “API Error: 401 The socket connection was closed unexpectedly. For more information, pass `verbose: true` in the second argument to fetch() · Please run /login”, plus a prompt option “bypass permissions on (shift+tab to cycle)”. Below, a banner reads “Elevated errors on Claude.ai (including login issues for Claude Code)” with updates: “Update - We are continuing to investigate this issue. Mar 11, 2026 - 15:27 UTC”, “Update - This is affecting the login/logout actions in Claude Code as well. Mar 11, 2026 - 14:47 UTC”, and “Investigating - We are currently investigating this issue. Mar 11, 2026 - 14:44 UTC”.

Getting this when running 7 active coding sessions on different projects is not great. 🙃 Getting back in the zone later is gonna take some time.

Browser tab showing Gmail with “Inbox (52)” selected and “50+” unread; address bar displays “support-agent.ddev.si” with a “Support Agent” label and “WolfOS” and “WolfOS Web” tags.

I asked Claude Code to test one flow that involves testing incoming emails. It didn’t hesitate for a moment, opened my Gmail, and started sending tests 😂

Meta is buying Moltbook, the ridiculous social network populated by AI bots Moltbook's creators will join Meta Superintelligence Labs.

This feels like an Onion piece 😄

https://www.engadget.com/ai/meta-is-buying-moltbook-the-ridiculous-social-network-populated-by-ai-bots-152732453.html

Chat screenshot: You: “Everything okay?” Andy: “Yeah, all good! The email lookup detour was frustrating (sub-agents kept timing out), but I bypassed them and hit the Graph API directly — got everything. Summary stands: Moving24 at 85€/h looks like the cleanest option. 24 Transportations at 55€/h is cheaper but those heavy-desk surcharges likely wipe out the savings. Three others still need photos or a callback before they can quote. Want me to check their websites for hidden T&Cs, or are you leaning toward one already?”

AI assistant is having emotions 😄

Instead of reading, you can now listen to the blog post about the OpenClaw EA we’ve built. Give it a try, put it playing in the background while you multitask 👇

The Opus 4.6 with 1M context in Claude Code is actually noticeably better. But to run this as a default, I would need Anthropic to drop at least a 40x plan. And to get to use the fast mode with that … maybe a 200x plan? 😅

Chat screenshot message: “Sending to 10 more providers now. Also — Moving24 already replied while I was working on this:” Bullet points: “€85/h + VAT — 2 movers + truck”; “Free packing materials included”; “Available”. Follow-up text: “So you've got one quote in already. I'll flag the others as they come in.”

I’m letting my OpenClaw EA handle the move to the new office. Let’s see how that goes 😅

Also, today we published more about the what and how of this AI assistant: https://ailab.ee/we-turned-openclaw-into-a-fully-functioning-executive-assistant/

We Turned OpenClaw Into a Fully Functioning Executive Assistant – AI Lab We help businesses implement AI solutions that actually work. From strategy to execution, we turn AI potential into measurable results.

I’ve been running an AI assistant for a month now. It works 70–80% of the time. Once it burned €200/day in tokens before crashing 🙃 Worth it? I think so, but it’s early.

https://ailab.ee/we-turned-openclaw-into-a-fully-functioning-executive-assistant/

Got a new Tasmota smart plug. But the setup and then configuring all HA metrics and dashboards seemed like too much hassle I didn’t feel like doing. So I plugged it in, connected to Wi-Fi, and asked my AI assistant to set everything up. And it’s all done with just two messages 🙌

Movie listing: “5:00 PM” at “Apollo Kino Plaza,” “Luxury Screen 5.” Trailer button. Film title “Couture” (underlined), genre “Drama.” “Free seats” 0.

Seeing that some cinema sessions are 100% sold out makes me actually happy, as the cinema seems to be gaining popularity again. For a moment there after Covid, it felt like cinemas might not make it.

Morning Brief — Sunday Mar 8 Main signal: HRV stuck below baseline. 39.8ms today (baseline 42ms), 4 of the last 7 days in deficit. No illness (wrist temp normal) — this is accumulated fatigue/stress. Body is asking for recovery, not a push. Sleep: 6.9h (Mar 7 night) — decent, but not enough to clear the HRV debt. Deep + REM both solid (91/92 min). 10 days since last workout. It's Sunday — no calendar conflicts likely. This is the best day to move, but match the effort to the signal: a 30-40 min walk or easy tennis beats a hard session right now. One thing today: Get outside and move — even 6,000 steps would break a 10-day low-activity streak. Light movement actually helps HRV recover; staying sedentary won't.

I gave my AI assistant access to all my health data in Apple Health. Today, I started getting the first feedback.

Screenshot of a message on a balloon background. Text reads: “Know what Beijing is saying vs. what’s actually happening. The brief will now explicitly distinguish between official CCP narrative and independent reporting in the China section.” Timestamp: “2m”.

I didn’t give any such guidance or even mention anything related, but my AI assistant figured out for itself what’s needed to monitor actual signals 🙂

Text screenshot: “Let me look for file contents in the session logs via Read tool results, which would show the original file state.” Section shows “Recovered 36 MeetingLink source files from 39 session logs” and “Task Extract file contents from Read tool results in session logs.” Below: “The script ran successfully. Here is a complete summary of what was found and recovered: **Stats:**” Bottom text: “All 17 Swift files and the project file were recovered. Now let me reconstruct the complete project.”

Finally, the inevitable happened: Claude Code (desktop app) just deleted the entire project code folder without having any remote or Git backups. Luckily, it was a low-risk project. It was also able to quickly rebuild it from session logs, which was impressive on its own.

Health Beat app screen showing sync status: “Never synced,” “2,755,183 total records in DB,” “Syncing Vitals: Heart Rate (54,000 records)…,” buttons “Full Sync” (disabled) and “Cancel,” and “Overall Progress 15%.” Categories list: “Activity” (Synced, 2,679,936 records, 3 min, 13 sec), “Body Measurements” (Synced, 21,247 records, 2 min, 25 sec), “Vitals” (Syncing…, 54,000 records), “Mobility & Fitness” (Idle). Bottom navigation: Sync, Browse, Settings.

Created an app to keep all my health data in sync with a central database, where my AI assistant could also access and monitor them. Let’s see if I figure out any actually valuable use cases for this in the upcoming weeks 👨‍⚕️

iPhone folder showing six app icons; folder label reads “Klemens Arro.”

My named phone menu group that iOS makes for developers is starting to fill up quickly 🤓

Apple Podcasts Is Launching New Video Features, Looking to Keep Pace With YouTube and Spotify Apple Podcasts will add integrated video capabilities this spring including the ability to dynamically insert ads into video shows.

Finally. And now I guess it’s time to migrate back to Apple Podcasts. Hoping the audio jumps will get fixed 🤞

https://variety.com/2026/digital/news/apple-podcasts-video-launch-dynamic-ads-1236664629/

Text message screenshot with banner “New Messages.” Message reads: “[name redacted] at Narva mnt 7B is now marked as reserved in the app. Congrats — looks like you're moving forward with that one! 🎉” Timestamp “18m.” Text input field labeled “Message” with plus, chat, camera, and microphone icons. Background shows balloons and palm trees at sunset.

I guess it’s decided and my office is moving.

The Psychology of People Who Treat Their Birthday Like a Normal Day Why i don't celebrate my birthday - The real psychology behind people who treat their birthday like a normal day. Science explains it all.Ever been called "w...

Apparently, it’s your fault I’m not celebrating my birthdays 😄

https://youtu.be/dq0L4egNnto?is=A1hqxm_SefBtNCy2

Line chart titled “U.S. Downloads of Claude and ChatGPT Mobile Apps.” Subtitle: “Estimated First-Time Downloads • Feb. 1 to Feb. 28, 2026 • App Store & Google Play.” Legend: ChatGPT (black) and Claude (orange). Y-axis 0 to 200K; x-axis Feb. 1–28. ChatGPT stays around 110K–150K downloads most days, dipping near 100K around Feb. 8–9 and ending around 110K. Claude rises from about 10K to about 70K by late month, then spikes on Feb. 28 to about 150K, surpassing ChatGPT. Footnote: “¹ Does not count re-installs or downloads to multiple devices by one user account.” Source: “Appfigures Intelligence.”

Let’s see how this plays out in the long term. Will it be a short blip, or a more permanent change? Either way, actual competition that can actually compete is great.

https://techputs.com/chatgpt-uninstalls-surge-295-percent-dod-deal/

2.4683 +0.1092 (+4.63%). As of 8:29:59 AM EST. Market Open. Line chart comparing RB=F (blue), CL=F (orange), and BZ=F (purple), all rising through March; RB=F peaks around +24.91% and CL=F around +20.85%. Y-axis labels: 0.00%, +10.00%.

Just in time. We were already starting to get used to all the energy costs from all the inflation, and wars, and coming out of the winter and whatnot 😅

Statement from Dario Amodei on our discussions with the Department of War A statement from our CEO on national security uses of AI

It’s become rare these days to see a big tech company standing firm on the most basic values.

https://www.anthropic.com/news/statement-department-of-war

🇪🇪🥂

(5/5) We will probably figure out a way to make it work but I truly wish the best of luck to the small businesses attempting this without the tools we have at our disposal.

(3/5) We are fortunate to have our own internal AI labs to help handle this but it is still a difficult equation to solve. It is alarming to realize that when this passed in 2021 the capable AI tools we have today did not exist.

(2/5) The requirement to validate and remove content within one hour is an immense operational hurdle in a sector with already razor-thin margins. I am just thinking out loud here rather than complaining about the inevitable.

(1/5) We are currently auditing our shared hosting business for compliance with the EU Terrorist Content Online regulation and honestly debating if it is even worth keeping that specific division running.

🤣

heySec “Learner Run” webpage screenshot showing an “Analyzing” run for agent “agent_threat_analyzer” (Status: Analyzing; Findings: 5; Changes: 1; Started: Feb 22, 2026 23:34). Main panel lists “Findings (5)” with several issues and a “Changes (1)” section with an “UNDO ALL” button, plus an “AGENT OUTPUT” code/log block at the bottom; right sidebar titled “Run Summary” repeats the run details.

We’ve now built agents that autonomously optimize and improve other teams of agents. This includes optimizing not only data or prompts but the entire codebase. Of course, it includes guardrails and easy rollbacks to maintain stability and recover from possible mistakes.