The 1000x engineer creates 1000x the output. What nobody seems to be paying attention to: where does the value of the other 999x go? Gains that large used to reach consumers as lower prices or workers as higher wages. For the last 50 years, neither has happened reliably. AI scales the production. Nothing in the current arrangement scales the distribution.
The "replace an entire team" framing is the bit that doesn't survive contact with daily use. I run an AI assistant across three businesses, and the honest version is that these tools replace tasks, not teams. ChatGPT and Claude are genuinely good at drafting, summarising, first-pass research. They're useless at the thing a team actually does: judgment, holding context across weeks, knowing when the obvious answer is wrong, and owning the outcome when it goes sideways. A list of 100 tools is a tab-hoarding problem, not a productivity one. The teams shipping fastest with AI aren't the ones with the most tools, they're the ones who picked three, wired them into a workflow with verification and a human sign-off, and ignored the other 97. A tool without a process is just a faster way to produce confident nonsense. Save the list by all means. Just don't mistake the bookmark folder for a strategy.
In all these AI discussions, what I would like to see is unbiased, well-reasoned thoughts on the ceiling/limit of the current ways of doing AI (causal transformer + backprop + RLVR + inference-time tokens). It's easy to daydream about infinite possibilities when assuming technology has infinite capabilities. But insights on ceilings/plateaus allow us to put real capabilities within scope.
One thread ties the entire conversation together for me: AI collapses the distance between intention and artifact, then pushes the hard problem upstream.
Once code, documents, simulations, regulatory work, and internal tools become cheap to generate, the scarce work moves to choosing the right target, designing the verification layer, and owning the consequences of the output.
That is why the “very large number of small teams” point feels so important. AI changes the minimum viable size of ambition. A founder can now assemble a factory around a problem before an incumbent has finished staffing the meeting.
The new leverage belongs to people who can move between domains, set the tests, and stand behind what ships.
One of the most insightful takes on AI I've read recently. The shift from "doing the work" to "building systems that do the work" is already happening. The biggest advantage won't come from using AI occasionally, but from learning how to direct, verify, and scale it effectively.
The software factory framing hits differently when you zoom out to industries that were never really thought of as software businesses but quietly became ones.
Digital advertising is a good example. What used to take a team of specialists to build, traffic, and optimize across channels now gets scaffolded in a few hours with agents handling the repetitive architecture. The 1000x engineer Naval describes is showing up in unexpected places, not just in dev shops.
Amazing read!
Thaaanks! It's amazing to really get the recent insights from Naval.
The 1000x engineer creates 1000x the output. What nobody seems to be paying attention to: where does the value of the other 999x go? Gains that large used to reach consumers as lower prices or workers as higher wages. For the last 50 years, neither has happened reliably. AI scales the production. Nothing in the current arrangement scales the distribution.
Unbelievable. Love it
@naval did u any AI to write contents. If yes, please give us some tips
The "replace an entire team" framing is the bit that doesn't survive contact with daily use. I run an AI assistant across three businesses, and the honest version is that these tools replace tasks, not teams. ChatGPT and Claude are genuinely good at drafting, summarising, first-pass research. They're useless at the thing a team actually does: judgment, holding context across weeks, knowing when the obvious answer is wrong, and owning the outcome when it goes sideways. A list of 100 tools is a tab-hoarding problem, not a productivity one. The teams shipping fastest with AI aren't the ones with the most tools, they're the ones who picked three, wired them into a workflow with verification and a human sign-off, and ignored the other 97. A tool without a process is just a faster way to produce confident nonsense. Save the list by all means. Just don't mistake the bookmark folder for a strategy.
In all these AI discussions, what I would like to see is unbiased, well-reasoned thoughts on the ceiling/limit of the current ways of doing AI (causal transformer + backprop + RLVR + inference-time tokens). It's easy to daydream about infinite possibilities when assuming technology has infinite capabilities. But insights on ceilings/plateaus allow us to put real capabilities within scope.
🙏👏
https://the3amnetwork.substack.com/p/what-netflix-wont-let-you-measure?utm_source=app-post-stats-page&r=lgym7&utm_medium=ios
The key isn't just using AI tools, but building judgment and creativity around them. AI can accelerate productivity, but it can't replace the human ability to make strategic decisions and innovate in complex domains. This framework may be worth clicking through to see how it applies to investing strategies. Check this out, it relates! https://fungalstockecosystem.substack.com/p/why-i-stopped-optimizing-for-cagr?utm_source=substack&utm_medium=comment&utm_campaign=engagement_copilot&utm_id=sec_8809cda0ff5f4b19&utm_content=soft_link_reply
Slopware alert
Awesome.
One thread ties the entire conversation together for me: AI collapses the distance between intention and artifact, then pushes the hard problem upstream.
Once code, documents, simulations, regulatory work, and internal tools become cheap to generate, the scarce work moves to choosing the right target, designing the verification layer, and owning the consequences of the output.
That is why the “very large number of small teams” point feels so important. AI changes the minimum viable size of ambition. A founder can now assemble a factory around a problem before an incumbent has finished staffing the meeting.
The new leverage belongs to people who can move between domains, set the tests, and stand behind what ships.
Mind-opening! And I learned many more new things from opening the links.
One of the most insightful takes on AI I've read recently. The shift from "doing the work" to "building systems that do the work" is already happening. The biggest advantage won't come from using AI occasionally, but from learning how to direct, verify, and scale it effectively.
The software factory framing hits differently when you zoom out to industries that were never really thought of as software businesses but quietly became ones.
Digital advertising is a good example. What used to take a team of specialists to build, traffic, and optimize across channels now gets scaffolded in a few hours with agents handling the repetitive architecture. The 1000x engineer Naval describes is showing up in unexpected places, not just in dev shops.
The idea of building your own AI factory is one of the most compelling mental models right now.