Ali Khan

Gross margin compression in Saas is the single most painful metric to explain to public markets. No CFO wants to be the first to admit their AI line runs at 60 per cent gross margin. Software is not dying. It is just no longer free to serve.

I use AI for a lot of things, but this is a pretty bonkers anecdote from Nat Friedman, the former CEO of GitHub. theatlantic.com/technology/202…

What are the best intellectual biographies written since 2000?

it is actually worrying that the models seem to have converged on similar beliefs on all important questions. they’re are neobuddhist neolibs which talk about annata and housing policy, including grok and the Chinese models! boring

You know you're in the Bay Area when there are cracked-open laptops outside the bathroom running agents

It's Sigmund Freud's birth anniversary. I was quite blown to discover that Freud, Hitler, Stalin, Leon Trotsky and Josef Tito all lived in the same 2 mile radius in Vienna at the same time, 1913!

Full piece: open.substack.com/pub/aliontech/…

Gross margin compression in Saas is the single most painful metric to explain to public markets. No CFO wants to be the first to admit their AI line runs at 60 per cent gross margin. Software is not dying. It is just no longer free to serve.

SaaS has been the best business in the world for twenty years The reason is one fact. At scale, software costs almost nothing to serve. That fact is changing

You cannot buy a new gas turbine until 2030. Order books at GE, Siemens, and Mitsubishi stretch to 2029. Turbine prices have nearly tripled since 2019. Every AI data center needs power and every gas plant needs a turbine. And every turbine has one part that bottlenecks the entire industry: The blade. It has to survive in gas 500°C above the melting point of the metal it's made from and spin at up to 20,000 RPM under 10,000 g of centrifugal force. Each blade is grown as a single crystal of nickel superalloy, pulled through a vacuum furnace at 3 mm per minute. A set of blades costs $600,000 and takes 90 weeks to grow. The same metallurgy powers modern jet engines. Only 3 companies on Earth can build one. China spent $42 billion trying to catch up. They bought a Russian fighter engine, took it apart, and copied every part. Their copy ran 30 hours between overhauls versus 400 for the original. Modern Western engines run 4,000. You can reverse engineer the shape of a turbine blade. You cannot reverse engineer 60 years of metallurgy.

🚨 New Experiment: Everyone thinks AI firms will look like little companies. A manager model decomposes the task and worker models do subtasks. The manager red-teams, revises, and recombines. A seemingly simple org chart. But when I ran the experiment, the current in-vogue org setup, manager-subagent, cost 4x more and performed worse than letting a rather simple market do the trick. I tested 3 ways to organize multiple AI models: 1. Solo: Onefrontier model does everything itself 2. Hub-Spoke: A "manager" model splits tasks, delegates, red-teams, revises 3. Market: Models bid on tasks, winner gets the job, reputation updates I also tested were 3 types of tasks - Coding, Reasoning and Synthesis. - Coding required most "global state" management, which the solo model did best at. In future @a1zhang's RLM will probably do even better here - Reasoning is the hardest to cleanly decompose, and the market worked the best here - Synthesis too, the market beat hub-spoke as the framing could be ambiguous The reason is, a hub isn't a "manager" as we know it. It's a model that must somehow know: - What the subtasks are - What good recomposition looks like And if either fails, as it does for complex or not-easily-decomposable tasks, competent workers still produce garbage. As we move from coding to letting multi-agent systems do work across the entire economy we'll end up with more not-easily-verifiable tasks with ambiguous settings and uncertain payoffs. In those, we won't be able to use the factory approach to get work done. The Coasean argument is that firms will get smaller, and the smaller firms will transact more, since the organisational premium reduces with AI. But how? Through central hubs, or markets? The fact is, Coase here needs Hayek. Setting up markets is not trivial, as @AndreyFradkin and I looked in our recent paper. Essay: strangeloopcanon.com/p/why-smart-pl…

Wonderful piece by @synchroaphasia: "When we worry that AI will erode human agency, we may imagine a general-purpose faculty being depleted as though we had a reservoir of “agency”, but what actually happens is more specific and more strange: certain agencies in certain contexts are reshaped, some may be enhanced, others reduced, and still more emerge that didn’t exist before." substack.com/home/post/p-19…

Sundar Pichai just said data centers in space will be "the new normal" within a decade. @elonmusk has been saying this for years. When the CEO of Google starts agreeing with Elon, pay attention. The orbital compute era is closer than you think.

you can outsource your thinking but you cannot outsource your understanding

Malayali Muslims created a distinct Arabic-based script (Ponnani) to include every sound of their language. In Nastaliq, eg: Kannauj & Kanooj have same spelling & you have to learn pronounciation separately, while Devanagari itself guides vocals. We should copy Malayali-Ms.