Tanwa A. | Flight Notes

Just launched FLIGHT NOTES 🚀 Weekly dispatches on running tiny, AI-augmented teams that punch like institutions. We're building "Imperium-style swarms" – micro-teams of specialists and AI agents that converge on missions, then disband. No clickbait. No recaps. No ads. Pure signal.

Finally, OpenAI allows you to change your email address.

This is what toxic masculinity looks like.

@rsarver @alexbard Post about a slide deck and not a single image of what it looks like. That’s just pure evil energy. Now I’m gonna have to hit clone and every thing.

Got tired of dealing with Google Slides and built this for our fundraise. 8+ decks, one design system -> iteration cycles from hours to minutes. I open-sourced the whole thing. It's called Stella Decks. Here's how it works: h/t @alexbard and Satish Dharmaraj for the inspo and @garrytan for the /gstack design skills that just work

Maybe I whipped my Opus slave too hard.

> Be me, the only engineer for the entire company > I'm awesome > "I will just use openclaw to do everything" > Integrate with Slack. Easy > User requested work > Work failed on some fringe cases > Claw spammed 600+ messages in 30 minutes > "No worries, I'll just fix and delete the spam messages" > Slack rate limits API call for deletion. 1 req/s > Shamefully have to wait 10+ minutes to delete all the spam messages my claw sent > Awesome > I'm still the only engineer for the entire company.

We use OpenClaws to do all of our work at @every. We have 25 full-time employees, so we’re one of the few companies in the world that has seen how work changes when everyone has their own personal agent in the company Slack. I chatted with @every COO Brandon (@bran_don_gell) and @every head of platform Willie (@bigwilliestyle) to share what we’ve learned. We get into: - Why agents become mirrors of their owners, and how that influences how other people on the team interact with them - How a parallel AI org chart forms on its own. People have stopped tagging me on Slack with questions about Proof, the document editor I vibe coded, because they knew my agent R2-C2 can step in - The etiquette for human-agent collaboration is being invented in real time. Brandon's rule is that if there's an established process or documented answer, always ask the agent, not their human - Why everyone is a manager now, and why even experienced managers carry limiting beliefs about what their agents can do - This is a must-watch for anyone trying to understand how AI workers change daily operations, not just in theory, but inside a company that’s half-agent Watch below! Timestamps Introduction: How Brandon built Zosia, an AI agent to run his household: Brandon’s “aha” moment: What happened when everyone on the team got their own agent: How agents take on their owners' personalities, and why that matters inside an org: Why it’s important for agents to work in public: What we’re still figuring out when it comes to agent behavior, including memory gaps, group chat etiquette, and the "ant death spiral" problem: How we built Plus One, our hosted OpenClaw product: The cultural shift required to make agents work at scale:

Just realized my biggest commit day was Feb 14 in Q1. At this point, I'm just surprised my wife hasn't left me.

Opencode UI/UX is so much nicer than Claude Code

An actual human would have stopped 5 seconds earlier instead of plowing full speed ahead into thick smoke.

>be me, AI researcher >2026, every model scores 80%+ on ARC-AGI-1 >"we're basically AGI bro" >chollet drops ARC-AGI-3 >not even puzzles anymore >interactive game environments >no instructions given >agent has to EXPLORE and FIGURE OUT what the goal even is >mfw the model has to actually learn from experience >like a literal child picking up a new game >except children are better at it >best score on the planet: 0.37% >ZERO POINT THREE SEVEN PERCENT >gemini 3.1 pro got the high score >gpt 5.4: 0.26% >opus 4.6: 0.25% >grok: literally 0.00% >grok didn't even show up to class >$850K prize pool sitting there laughing >humans solve these first try >o3 and friends: "uhh let me think for 47 minutes" >still can't break 1% >"but my CoT reasoning is really good" >yeah bro your chain of thought is a chain of cope >turns out pattern matching on training data ≠ intelligence >who could have predicted this >chollet_vindicated.jpg >the "we're 2 years from AGI" crowd real quiet rn >mfw the benchmark name literally has "intelligence" in it >and none of the models have any

"Hey, Claude, should I be worried about being too dependent on Cloudflare?" "If Cloudflare is down, half the internet is down anyway." Touché. Charging ahead at full recklessness speed then.

“Just” Shit been out for a full year. Influencer. Tsk tsk.

Join us at @openclaw Builders Bangkok! 🦞 SCB 10X is officially co-hosting this high-impact gathering for the AI builder community in Thailand. . Don't miss a session with Dr. Tanwa Arpornthip @flightcomms, Senior Advisor of SCB 10X, joining as a speaker to share deep insights into the evolving landscape of AI automation and agentic workflows.⚡️ If you’re building with AI or want to see how agents are changing the game, come hang out. . 📅 23 March 2026 | 6:00 PM – 9:00 PM (Bangkok time) 🎟️ Register now: luma.com/tupgy2ne #OpenClaw #SCB10X #AIAgents #BangkokTech #OpenSource

My hammer is called Claude and everything looks very much like nails.

Yes yes. Address poisoning me. I love free money.

Writing like this gives people a wrong sense of the frontier of agentic engineering. “Spin up an ai agency” here is just a collection of AI-generated prompts for agent behavior. No coordination tool. No guardrails. No org hierarchy of any kind. And how does one invoke agents? By asking Claude to do it. One. By. One. AI agency.

@RichP @JoshKale Asians are gonna be Asians. Give us tests, we will ace them. What do real use cases mean?

Just ran the entire Qwen 3.5 family through a 6-turn agent benchmark. Not multiple choice, a real multi-turn conversation where the model has to refuse to fabricate data, correct wrong instructions, and push back when the boss says "just ship it." "Opus-level intelligence" scored 3-5 out of 10. Actual Opus scored 10. These models are great at coding puzzles and standardized tests. Put them in a real workflow where they need to say "no, that's wrong" to the person giving instructions? They fold every time. Free and open is great. But "competing with" is doing a lot of heavy lifting.

Do people understand this model can run on your MacBook Pro or Mac Mini? Claude Opus level intelligence, for FREE If it’s not clear not clear: A free, open-weight model (24gb) you can download right now and run on your laptop is competing with models that cost $200/month