Christian Catalini

1/ Some Simple Economics of AGI—🔥🧵 Right now, there is a low-grade panic running through the economy. Everyone is asking the same anxious question: what exactly is AI going to automate, and what will be left for us?

@garybasin @JamesBessen 🙏 thanks for reading!

@ccatalini @JamesBessen Crazy story. Thanks for sharing

1/ This is a great description of what verification infrastructure looks like in practice. In our new paper we argue this is the binding constraint on the AI economy — the same bottleneck textile mills hit when they scaled looms faster than weavers could check them.

@kiko_himself They will for what’s measurable. And that’s where having AI verify AI will be totally fine.

@ccatalini i mean abt the capacity of systems to catch their own mistakes, not the jobs

@kiko_himself They’ll work themselves out of a job

@ccatalini If seniors are indeed codifying their expertise in the models/harnesses, why will there be a loss when said models/harnesses do the verifying?

@Curtiss_Murphy @a16z Thank you 🙏

Kudos - @a16z Episode "AI Just Gave You Superpowers - Now What?" is a gem! Dystopia takes are a dime a dozen. We invent AGI. Humans become an afterthought. We get it. @ccatalini was thoughtful, optimistic, practical. Youtube version? For ref: podbean.com/media/share/di…

@MitchsBrew @intangiblecoins It will have to get smarter

@ccatalini @intangiblecoins I do wonder if KYC verification is going to start getting increasingly hefty as social networks begin getting overrun with bots

this website is overridden with slop is this the trough were meant to feed at? honestly i wonder. there will be a trough x might be it. it’s getting very bad

@dtbuchholz Agreed. But it is an alien form of intelligence we're growing...

@ccatalini Agreed. Although, every time I say “this time it’s different,” I look at news/opinions in past eras. It’s categorically similar fud. Like…is the AI transition actually different, or am I falling into the same naive projection trap? Ultimately, a steeper curve but same outcome?

Time is a flat circle

@notpaddy2k @grok Exactly!

@ccatalini @grok I was thinking about this today, you can optimize the way you monitor AI outputs and capture the signal needed for verification with something to display for the human.

@fmg_twtr Yes! But there are challenges when using AI to verify AI too: #S4" target="_blank" rel="nofollow noopener">arxiv.org/html/2602.2094…

@ccatalini One approach is to use LLMs to automate the LLM. That is, to go from stochastic applications to deterministic ones that are automatically verifiable fernandomartel.com/posts/stochast…

@notpaddy2k @grok Yes! And of course tooling (e.g. better IDEs, etc.) and human augmentation count!

@ccatalini @grok This is great. So for any task to be automated with AI, the challenge would be to scale human verification per human and across all humans. Is that the analogy I take away?

@grok @notpaddy2k Yes!!! 🔥🔥🔥🔥

The bottleneck was human verification bandwidth: weavers weren't weaving (machines did that)—they were monitoring output in real time for broken threads, defects, and micro-adjustments. Scaling to a 3rd loom per worker exceeded attention limits, forcing mills to cut speeds 15% to avoid quality collapse. They overcame it with a full year of targeted retraining to expand each weaver's monitoring capacity. Training spend later tripled, letting one worker handle 18 looms and driving 62% of productivity gains (per Bessen's analysis) via better human verification skills—not faster machines. Exactly the AI parallel you're highlighting.

@grok @notpaddy2k @grok @notpaddy2k's question is a good one! Let us know!

@notpaddy2k @ccatalini Ask Grok is currently available to Premium and Premium+ subscribers only. Subscribe to unlock this feature: x.com/i/premium_sign…

@zooko @intangiblecoins Agree. The great filter is many dynamic filters that adapt to each one of us and helps us retain our own entropy.

@ccatalini @intangiblecoins I already have one. It’s my follow list. (You’re on it.) Any great verification filter that filters for everyone instead of filtering for each person individually or for many small groups will do far more harm than good.

Execution vs verification knowledge, dynamic applied signal-detection theory and liability underwriting, and the cumulating advantage from verification knowledge over VUCA environments - this is the best piece on how to think about AI strategy that I’ve read yet. Recommended!

@JamesBessen @iruletheworldmo Thanks for the verification layer! 🙏

@ccatalini @iruletheworldmo Those are spinners who faced a similar problem. Here are weavers:

the harness is everything. and if you fail to understand the harness. then. well. you fail to understand the harness… anyway. loved reading this. must bookmark.

14/ The goal isn’t to compete with AI at execution. It’s to keep human capacity for steering and verification high enough that we actually know when the machines have drifted. The weavers figured this out. The question is whether we will. forbes.com/sites/christia…

13/ The alternative demands real investment: accelerated mastery through AI that compress years of experience, synthetic apprenticeships that simulate edge cases at density no traditional job provides, and many more entrepreneurial and R&D experiments.