There are a number of incredible low-margin consumer services that have massively grown their operating profit (and their valuation) through an in-product advertising business: - Amazon: ~$80B - Uber: ~$2B - Doordash: ~$1B - Instacart: ~$1B It'll be interesting to see if this creates a modern form of the innovator's dilemma, i.e., enabling agentic use (which is unequivocally pro-customer) cuts deeply to their bottom line, creating opportunity for those that don't rely on ads for their operating profit (e.g., Shopify).

A great regret is that I did not overlap much with @moritzKBE. His internal writing and oratory was just as insightful, incisive and enjoyable. Once (and I think only once), a meeting landed on my calendar with MM as host. I had zero context. I joined him and we sat down for an hour, opposite an individual (a founder? recruit? who knew?). The next 60 minutes was exactly as advertised. A charming, detailed exploration of the person's background and childhood. I think by the end of the meeting, we had arrived at the person's entrance to adulthood. We warmly wrapped the meeting and walked back to our desks. I was still confused but suspected this had been a founder meeting. Nothing (AFAICR) was uttered about his present company or its prospects. As we sat back at our desks, MM turned and said "well, that was pretty interesting, wasn't it?" One of my core lessons from venture is a bastardization of the opening of Anna Karenina. Mediocre investors are all alike; every great investor is great in their own way.

I am biased but there are some pretty good companies on this list to think about joining: - Instinct is building the personal personal assistant - @KeenableAI is building the real-time search engine for agents, built by a cracked ex-search team - @quadrillionlabs is building the platform for auto-research - @trajectorylabs is building the platform for continual learning - @watneyrobotics is low-key managing all those AI data centers you keep hearing about You should think about joining them! All have <25 employees.

One of the crazier facts about venture is that it seems consistently possible to build a small community of college-aged kids that is incredibly talent dense, e.g.: - Thiel Fellows (ETH, Figma, I think Cognition) - KP Fellows (Opensea, also Figma) - Dorm Room Fund (Cursor) - Prod (Mercor, Etched, also Cursor) - Neo Scholars (Cog, also Cursor) Other than Neo (@neo) though, this hasn't really become anyone's dominant strategy.

Pranav (@pranavreddy): "I asked Claude to tell me how good a writer I was on a scale of 1-10. It kept refusing, insisting that my question came from a place of insecurity and it was unwilling to judge or evaluate me on this premise." Me: "I wonder what it would say if I asked?" Claude: "Ah, writing — that I can actually rate. Based on what I've seen: a solid 7." There were no winners in today's game.

Legato is a super cool product. It's like the Meta Displays but w/o the at-times-awkward camera. It's initially focused on those with hearing loss – it's a lot more subtle than an in-ear hearing aid – but it also works great for calls and hands-free AI usage. Congrats @mehuligan & team!

I thought this part was particularly interesting: How can we know that two engines are dependent? It is easy to measure. Engines converging on relevant results proves nothing. What they shouldn't share is mistakes. When two engines keep returning the same irrelevant documents for the same queries, the simplest explanation is a shared upstream index. Two students with the same right answer studied. Two students with the same wrong answer sat next to each other. We combine raw overlap and shared-mistake rate into a single chart.

Perception: •⁠ ⁠This is the craziest founder reveal ever! •⁠ ⁠This is the most insane velvet-rope launch strategy ever! Reality: •⁠ ⁠Some people just prefer quietly building vs. tweeting •⁠ ⁠Compute is really scarce right now! Have you not been reading Twitter?

Instinct is a personal agent that works. There are lots of personal agents. But we think Instinct is special because it really works — we want it to be able to do everything that a human could: use a phone, browse the web, make a note, take a call. We want Instinct to work for everyone. There’s a lot of companies building “AI for Enterprise” (and we invest in lots of those too!) but there doesn’t exist a proactive, capable product that tries to AI for the rest of us. I met Noah for the first time last August. By reference, and by first impression, he was obviously exceptional: Noah is down-to-earth, quietly competitive, and absurdly focused. He went from an undergraduate at Northeastern to industry-moving agent research shockingly quickly. @mvernal and I knew immediately we had to work with him. When Noah texted that he was raising a round, I flew back in early to talk to him— we’re really lucky he picked us. For the last ten months, I’ve spent almost every other Friday at the Instinct office. The Instinct team is small and opinionated. It’s a group that takes the responsibility of building the world’s personal agent very seriously but with joy. The reception we’ve had is a reflection of the thought and care that the team has put into every detail of how Instinct thinks about and talks to you. We @conviction are so proud to work with Instinct and Noah, and excited for what Instinct will do for all of you.

Noah (@noahrshinn) and Instinct are both phenomenal. It's a really, really good product. We love it. And Pranav (@pranavreddy), who led the first round in the company, doubles as a very capable, WSJ-cited photographer!

One of the cool, non-obvious things about building your own index and AI-native search experience is you optimize for a completely different set of things. Human queries are a couple of words and the human almost never looks past the first page of results (or even just the first result). An AI might pass thousands of tokens as query context and might read thousands of results. Google and its modern wrappers end up with an intrinsic bias towards human query patterns which is hard to break unless you start from scratch.

At this point, pre-training basically consumes all extant human knowledge. @KeenableAI provides the rest – it is a constantly updating, versioned, AI-native index of human knowledge built by an insanely cracked team of search engineers (led by @styskin, the former CEO of Yandex Search). We are incredibly bullish this team and product. If you are interested in using it or helping to build it, please reach out and happy to connect you.

I cannot understand why @googlecalendar does not recognize a phone number in the location field and just link it to a phone call vs. sending me to Google Maps to search for a phone number. I want to go work at Google for three hours just to fix this.