(#151) Group chats as distribution: How OpenAI is disrupting messaging without the graph
WhereIsMyMoat.com is now live and already busy. 🚀
Last week I started a new project, a personal one related to investments. Here is what you can find new this week
If you enjoy my strategy notes on LinkedIn or in OnStrategy, WhereIsMyMoat.com is the deeper, “skin-in-the-game” version: full theses, moat scoring, and clear buy/hold/sell views.
Have a look, read the Apple sample, and if it’s useful, consider subscribing.
– Sorin
—
On to update:
How OpenAI is disrupting messaging without the graph
OpenAI’s rollout of group chats in ChatGPT is a small UX tweak with massive implications. On the surface, it’s a productivity tool for shared itineraries and brainstorming sessions. But underneath, it’s the perfect viral growth hack. It is a feature that invites users to bring their social graph into the product, without needing to build one. You don’t have to be Facebook; you just need a link. Every group becomes a mini-distribution node, a collaborative space that makes ChatGPT more useful the more people you add. And crucially, it embeds ChatGPT in everyday coordination and conversation flows, hence turning "use the AI" from a solo act into a default social behavior.
Group chats with embedded agents should have been Meta's playground to dominate. They had the graph, the network effects, the messaging platforms, and a cultural monopoly on "where group conversations happen". Instead, OpenAI ships it first. With no feed, no scrolling, and, for the moment, no ads. Just shared utility and a product experience that teaches users to loop ChatGPT into real-world decisions. This looks to me like a distribution disruption. Meta built the stadium. OpenAI just showed up with the ball, the scoreboard, and half the crowd (= 800 MAU). LINK
The Chinese EVs tsunami is here
Look, the global EV sales chart is basically a Tesla ad with some BYD seasoning and a few mystery entrants like “Changan Lumin”, which sounds more like a vitamin supplement than a car. Tesla Model Y is lapping the field with 140,904 units in September, and Model 3 holds second. And yeah, BYD has eight cars in the top 15, which is impressive if you grade on volume rather than design language. But let’s be honest, no one in Munich or Milan wakes up dreaming of a BYD Sealion. It’s the IKEA of vehicles: practical, inexpensive, and mildly confusing when you try to explain it to your dad.
Meanwhile, Germany is stuck in a philosophical debate about whether cars should be built from steel or soul. Everyone wants the feel of a Mercedes or a BMW, not a QR code with wheels. The problem is, while Stuttgart polishes its heritage and debates EU emissions credits, the rest of the world is buying cars with weird names, touchscreen dashboards, and range anxiety. If the next Porsche is going to compete with the Changan Lumin, someone in Germany better stop building “driving machines” and start building delivery machines, because plastic or not, the race is being won on scale and speed, not legacy and leather stitching.
China’s EV strategy: export to Asia and Europe
What that chart really shows is that “China” in EVs is no longer a country, it’s an aggregator of scale. Once you’re exporting 700k+ vehicles to Europe and a similar number around Asia, the learning curves on batteries, motors, inverters, and factory automation stop being linear and start being compounding. Europe can slap tariffs on the finished car, but it can’t tariff away China’s cost advantage on the stack: batteries, supply chain control, and process know-how. In other words, the unit that competes with Europe is not BYD or SAIC individually, but against the entire Chinese production system, totally backed by the Chinese state.
The European “capitulation” scenarios are exactly how this kind of advantage plays out in practice. First, Europe buys the brains (Chinese software, ADAS/”Autopilot”, vehicle OS) because that’s the easiest way for a legacy OEM to ship something that doesn’t feel 10 years behind. Then it rents the muscle (Chinese-owned or Chinese-run factories in Europe) to arbitrage tariffs and politics while still tapping into China’s cost structure and manufacturing playbook. On paper, Europe keeps the jobs and the brands; in reality, it becomes the front-end UX and regulatory wrapper for a Chinese industrial platform running underneath. Maybe the European premium EVs will have a chance. Just maybe. LINK
Any chances of having a European self-driving car?
You know things are bad when startup founders in the US start saying, with a straight face, “This company wouldn’t exist if we had started in Europe” [1] ---> not as humblebrag, but as a completely accurate economic diagnosis. Europe totally lost the plot on self-driving cars because it regulates like a nervous librarian, funds like a slow-motion bake sale, and then holds panels on why it’s falling behind. While San Francisco has Waymo and Cruise dueling for robo-taxi supremacy (and China has many more starting with Pony AI), Europe’s most promising autonomous vehicle deployment is still in “development” and probably delayed because Brussels needs three more committees to harmonize parking lot signage.
Now comes the part where some EU commissioner gives a speech in Turin about “tech sovereignty” and “AI gigafactories” while the rest of the world deploys actual, functioning tech at scale. The reality is that you can’t innovate at committee speed, and you can’t build transformative platforms when your national funding schemes cap out at EUR2.3 million and come with 900-page grant applications. Meanwhile, in the US, someone can raise $40 million off a Figma prototype and vibes. Europe’s strategy seems to be “wait until American/Chinese companies are so dominant they open a showroom in Munich, and then complain about fairness”.
Maybe someday Europe will get a real AI car on the road. Until then, enjoy the clean sidewalks and good win, because the future isn’t stopping there. a16z, Bloomberg
Who’s taking our jobs? This:
Someone on Twitter: “i can’t believe this is taking our jobs”
It is pretty funny that “this” is what’s allegedly taking our jobs, because “this” is basically a glorified autocomplete that read the entire internet and then got really good at faking confidence.
For years, the implicit social contract of white-collar work was the following: if you learn the right jargon and can rearrange bullet points in powerpoint at 11 pm, you are doing something mysterious and therefore valuable.
Now, LLMs break the spell by showing that a surprising amount of that “mystery” is just pattern-matching over emails and PDFs. The scary part is that a lot of our work is dumb in exactly the way a model can copy.
So when someone says “I can’t believe this is taking our jobs”, what they’re really saying is “I can’t believe my job was basically a well-paid autocomplete prompt”.
Recommended by LinkedIn
The remaining jobs are the ones where you have to pick the problem, own the risk, and be on the hook when things go sideways… at least until someone fine-tunes a model on blame-taking and board presentations.
For Elon, the hard part begins now:
Bloomberg: “Tesla Shareholders Approve $1 Trillion Pay Package for Musk”
Elon Musk just got shareholder approval for a $1 trillion pay package, which is the kind of sentence you expect to end with “…and then the SEC fainted”. But here we are: 75% of shareholders basically said, “Sure, take a shot at becoming the richest human in planetary history as long as you give us robotaxis and $400 billion in EBITDA.”
Honestly, if you’re a Tesla shareholder and you’re not in favor of this, I have questions. Because this is a contractual hallucination of value creation. Musk only gets the cash if Tesla becomes a $8.5 trillion company. That’s a performance art piece about optionality. If he fails, he gets nothing. If he succeeds, you’re rich too. Anyone normally insane should want this deal.
And yes, sure, Bernie Sanders is mad and Norway’s sovereign fund had feelings, but let’s be real. You don’t build Optimus and a terafab chip plant on vibes and committee governance. You do it with a guy who treats EBITDA targets like side quests on his journey to colonize Mars. The real risk for shareholders was that he’d get bored, tweet something insane, and disappear to build flamethrowers and humanoid drones full-time. This deal keeps him tethered to the machine.
And if Tesla hits even half these targets, everyone screaming about income inequality will be too busy riding in self-driving Cybercabs to care. Bloomberg
Key speaker at the "AI@WORK:HR" conference
This week, at "AI@WORK:HR" conference, I discussed how AI is also reshaping the Human Resource function
😎HR leaders will need 4 “super-skills” to act as a sort of “Chief Anxiety Officer” - reducing fear around AI, not amplifying it:
1️⃣ AI & data literacy - understand models, bias, dashboards
2️⃣ Product-manager thinking – employee journeys, funnels, test & learn, not just procedures.
3️⃣ Change leadership - managing the anxiety around AI
4️⃣ Deep human skills - coaching, ethics, judgment, that is, exactly what a model can’t automate.
On the question "Data vs intuition?" I answered that we should think in the pattern of "Data-first, human-final".
I like a 70/20/10 rule:
Thank you for the invitation, Mehdi Vlad Alishahi & Team!
Key speaker at the Chamber of Tax Consultants (Camera Consultanților Fiscali):
Key takeaway from my talk today at the Chamber of Tax Consultants (Camera Consultanților Fiscali):
1/ we’ve entered the "Compressed Decade" where a ten-year shift packed into the next 24-36 months. AI is collapsing the price of cognition, making software teachable in plain language, and turning every workflow into data.
2/ for tax and finance, that means continuous audits instead of annual ones, first drafts in minutes, not days, instant scenario modeling, better fraud detection, and small teams with enterprise-grade leverage.
3/ The winners won’t "experiment with AI", but they’ll productize copilots around their core processes, instrument everything for traceability, and keep humans focused on judgment, ethics, and edge cases.
4/ The strategy is clear at this point, ie. automate the routine, augment the expert, and compound the learning.