Welcome to Slow Ventures’ Snailmail, where we slow down and share what’s been on our partners’ and founders’ minds.
TL;DR:
Others:
Build AI Accenture, Not Accenture for AI
Yoni Rechtman
Context
We're in the installation period of the AI supercycle: hype, speculative capital, and technological progress all rampant. The deployment period, where AI becomes usefully integrated into our productive base, is just beginning. It's going to last generations.
Deployment is the most important thing happening in the world right now. Acceleration and abundance vs. stagnation rests in the balance. So it's not surprising that there's intense, widespread interest in betting directly on it, not just on the beneficiaries (new AI-enabled products and services) but on the delivery (the IT services firms working on AI transformation projects for legacy institutions).
System integrators and consultants are the connective tissue between installation and deployment. The question is what flavor wins.
Market Signal
There are three or four main flavors of AI services right now, and they aren't equivalent bets:
The bodyshops. Elite generalist FDE teams parachuting into enterprise transformation work horizontally and per-engagement. Rational for now — things move too fast to build product. Also a much better deal for a certain kind of talent, who'd rather be the hero than run customer success for a research team that sees deployment as overhead.
The captives / JVs. Deployment arms spun up by labs and hyperscalers. The math works for the sponsors: $100M to improve the odds for a $10B position is rational. At $1T it's necessary. The buyer's risk is a Hotel California situation — their AI gets built on someone else's stack, with someone else's FDEs, around someone else's models. Their people aren't working for your outcome; they're working for their sponsor's lock-in. Many buyers will take the deal anyway.
The product companies. Specialists building an internal product that delivers a specific outcome for a specific customer, repeatedly. We have one company here operating as a system integrator. We're looking for more.
The most interesting version of the story is a product company delivering a service, rather than a service company delivering a product. The internal product (agent/harness, context layer, workflow automation) enables differentiated service delivery. These companies bend the cost curve down, instead of charging more because of temporal buying pressure and access to scarce talent.
Takeaways
The tell is the ICP. Will you do anything for everyone, or do you have a narrowly defined ICP? Product companies will have clear religious beliefs about what they can't touch or won't do. Body shops won't.
Recognition isn't a moat. The JVs have max access and brand. That gets them in the door. It doesn't make them durable.
The opportunity isn't to deliver AI projects faster. It's to fundamentally rebuild how services get delivered. AI Accenture, not Accenture for AI.
The New X Algorithm: What Actually Works Now
Sam Lessin
Context
I've been running enough experiments on X lately that I now have a working mental model for how the new Phoenix retrieval ranker actually behaves and the practical implications for anyone using this platform to distribute ideas are kind of huge.
The TL;DR: long native posts win on reach. Screenshot essays win on in-network engagement. The combo wins overall. And it's not really debatable anymore, the data is consistent across enough posts that I'm now changing how I publish.
Market Signal
Phoenix retrieval embeds the text of the post to decide who outside your followers should see it. That means a screenshot (an image with no machine-readable text) leaves the retrieval system flying blind. Your network might love it (because they already follow you, they're in the in-network distribution pool), but the algorithm has nothing to grab onto when deciding whether to push it to anyone who doesn't already follow you.
The native long-form post is the opposite. The algorithm can read it, embed it, and figure out who's likely to engage with the ideas. That's how you reach the next ring of people.
But there's a wrinkle: screenshots still get better in-network engagement, because they look more "premium", they imply you put thought into this and it's portable, it's something you'd want to save. So the best move I've found is to post both. Native long-form for distribution, screenshot for the in-network dopamine loop. The combo wins.
This is the practical part of what I think most builders and operators are missing right now. If you've been treating X like a screenshot-first platform (and a lot of people have, because that's what worked in 2024) you're throwing away a meaningful chunk of your reach.
Takeaways
Long native > screenshot for reach. The ranker reads the text. No text, no out-of-network distribution.
Screenshot > long native for in-network engagement. They feel premium. People bookmark them.
The combo wins overall. I'm now defaulting to posting both for anything I actually want to land.
Our New Merchandise Store
Team Slow
In light of the Wall Street Journal homepage feature on our Etiquette Finishing School, “The Tech Bros Are Going to Etiquette School” we know we can’t bring the experience to every city that’s requested it (Tel Aviv, Washington DC, Niseko, and beyond). So we did the next best thing: launched a merchandise store that lets you wear the thesis instead. From bowties ($30) to bar soap ($15), you can now practice modern etiquette wherever you are.
We’ve also brought some of the industry’s most widely held but rarely stated beliefs into the physical world, so you can wear, gift, or leave them sitting conspicuously on your desk. Think: When Masa Buys, You Sell ($30) and the AI Bubble Igniter Kit ($15).
Limited quantities only. Cowboy hats sold separately.
More Musing From The Team