18 Comments
User's avatar
Tristan Camacho's avatar

Wonderful to know that this is all possible, but as someone who's currently trying to transition into operations, I'm concerned that this would wipe out the bottom rung of that career ladder. I'm assuming that the solution is to learn Claude Code, but I don't know how to code. Does anyone have any resources they could suggest so that I can upskill and stay ahead of the curve?

Expand full comment
Alex Petropoulos's avatar

what's cool is that Claude Code can teach you how to use Claude Code to some extent. I also found this thread from Boris quite useful once I was around mid double digit hrs into using it: https://x.com/bcherny/status/2007179832300581177.

Also, when you start up using it you can tell it to store in the claude.md file (it's memory for all instances) some instructions something like "I'm non-technical but still want to learn, can you explain things to me in a way that lets me use you as a non-technical user".

Expand full comment
Tristan Camacho's avatar

Programmers programming the program to program the programmers. Haha. Thank you for this!

Expand full comment
Henry's avatar

If you know how to ask for what you want, you know how to use Claude Code. Entry-level knowledge work roles have collapsed - partly due to AI and partly due to high interest rates. This is timely career advice: https://80000hours.org/agi/guide/skills-ai-makes-valuable/

Expand full comment
Tristan Camacho's avatar

Ah, yes. I know this article well. Thank you. The reason I'm trying to get into operations in the first place is because it's a stepping stone to getting into program management, which is what I want to do in the AI policy space. Admittedly, I have a lot of learning to do to get to that point, and it seems Claude Code will be another skill I'll have to add to the list.

Expand full comment
Cobus Kok's avatar

Great article! The two-tier economy won't be about who learns the terminal. It'll be about who can specify intent clearly. The bottleneck is shifting from "can you execute" to "do you know what you actually want." That's a much stranger skill gap.

Expand full comment
Steeven's avatar

There’s another important part…Claude code is so fun to use. You never get incorrect errors due to syntax issues or because you didn’t know the command or that a tool could solve your problem. You just keep going so now your ideas become real in time to make your initial interest in a concept enough to turn it into something that exists

Expand full comment
Andrew Bowlus's avatar

I’ve been using Codex CLI since I don’t want to pay $$ for Opus 4.5 in Claude Code. How much am I missing by using 5.2-Codex-Max instead of Opus 4.5?

Expand full comment
Rob Wiblin's avatar

Shakeel I've installed Claude Code now - can you suggest some non-coding activities/tasks that I and other readers could give a go in order to get an intuitive sense of how powerful it is as a knowledge worker?

Expand full comment
Shakeel Hashim's avatar

Travel planning

Give it a bunch of PDFs on ~any topic and ask it to synthesise them

Data analysis — give it eg your podcast stats and ask it to analyse it, find patterns, and produce a report

This is a coding one, but: think about any repetitive action you do on your computer and ask it to automate it.

Expand full comment
Shakeel Hashim's avatar

oh, and doing your taxes!

Expand full comment
PS's avatar

I've spent some time looking through the examples here and in the comments below (and discussing them with Claude), and am still a bit confused. It seems obvious that Claude Code makes work with local files easier, in particular if you want to actually edit/create something. But synthesizing PDFs or analysing podcast stats could be just as well done with Web/Desktop, no? Yes, you need to upload the files manually, and if you do it all the time it's probably a bit easier and/or faster to do it in Code, and the size limits might be better, but that isn't really a that much of a qualitative difference. And the same applies to claude.md - as far as I understand, it's just a different form of project instructions, which might be more a bit more useful in some cases and a bit more effort in others, but again, not really a game-changer. What am I missing?

Expand full comment
Shakeel Hashim's avatar

It's a fair point -- but something about the harness makes it smarter and more competent. It's less lazy, in a way, and more persistent and autonomous. It runs for longer and can do a bunch of tasks in sequence (or parallel), rather than one at a time. Hard to fully put your finger on it, but it gets better results.

Expand full comment
PS's avatar

Thanks! That's valueable enough, will give it a try. I was just trying to understand which benefits were due to the underlying model vs environment vs tooling. I feel that some articles were throwing these things together, but I know you and many others are much more thoughtful and knowledgable than me, so I probably just have difficultty grasping the concept...

Expand full comment
Shakeel Hashim's avatar

No, it's a very good question -- and it's genuinely hard to say which is the thing driving the capabilities!

Expand full comment
Nicholas Wagner's avatar

If you are interested in a basic finance project with Excel, I have a guide written up on my company website. https://www.learningjourneyai.com/learn/fpa-usage-based/

I would suggest data gathering and analysis as activities that used to involve coding but are handled well by Opus 4.5 in natural language, e.g. https://x.com/mattyglesias/status/2007944461930877172

Other uses:

* https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code

* Take a folder of meeting transcripts and use them to populate a sales CRM or do topic modeling.

* https://x.com/Hesamation/status/2007804124469817687

Expand full comment
Henry's avatar

Great points but conclusion feels like it was written by AI - *everyone* would notice if AI caused mass (>5%) unemployment?

Expand full comment
Shakeel Hashim's avatar

hah, that was all me! agreed that we’d notice when unemployment hits, but i think there will be a big lag between AI tools being capable of causing mass unemployment, and it actually doing so (and i think that lag time is going to be a really important period for action)

Expand full comment