If you feel like the tech people in your life and on your timeline have collectively lost their minds, but more so than usual, that’s just Claude Code’s experience at work.
Now, if you know what I’m talking about, you’re either encoding vibes so hard that you’re about to dissolve into digital ecstasy or you’re breaking out in a cold sweat and composing your “I, for one, welcome our AI overlords” email.
But if you think Claude Code sounds like a New York Times pun you haven’t tried yet, this FAQ is for you.
Okay, so what is it?
Well, do you know how chatbots…chat? Write to you, talk to you, write your university papers? Claude Code, which comes from AI company Anthropic, is an AI tool that can actually do things with your computer. Actually, many of the things you can do with your computer. (Well, not you, if you’re the target audience for this FAQ, but someone who is an expert programmer who never sleeps, never says no, and works at impressive speed.)
Do you… like what things?
Honestly, it would be easier to list the things you can’t do with a computer. But an incomplete summary of what users have achieved with Claude Code would include: a Spotify Wrapped program but for text messages; personalized daily digests including emails, newsletters and more; a Pokémon card management system; a personal DNA analyzer; and a “cyberpunk” Tetris game. You’ll need at least a $20/month Claude Pro account, with no freebies for you.
…Cool? But it has “code” in the name. Do I have to know something about programming?
Don’t worry! Yes, Claude Code is designed to work in what’s known as a “command-line interface,” or the part of your computer where, instead of clicking on icons or typing normal sentences, you type commands with a programming language into a terminal, also known as the black screen where nerds enter their code.
Whoa, whoa, whoa! Do I look like Angelina Jolie from the 1995 movie? hackers? I don’t know what any of that means.
Okay, me neither!
It’s true that experienced programmers can get the most out of Claude Code (although they’re also the ones going through the deepest existential crises). But the learning curve for using Claude Code is descending faster than a Six Flags roller coaster, and you can increasingly interact with Claude Code more or less as you would with a chatbot if you want: with plain English and relatively few commands. Note that it’s more complicated than using it in the terminal, but I honestly wouldn’t trust any of us with that.
In a nutshell, the process works like this:
- You tell it what you want (fix a bug or create a new feature).
- Examine the project’s code base (all the files that make the program you’re working on run, including the actual code plus the configuration and test files surrounding it) to understand what’s happening.
- Edit the relevant files.
- You can run tests/commands to see if you broke something.
- Iterate.
At best, it closes the loop, mostly on its own: plan → change → check → fix. That’s why people who make their living creating software act like they’ve freed themselves from thousands of little paper cuts.
But I would like to keep my files. Ideally, it should be all of them. In its current state of existence.
Intelligent person. Claude Code has an agency appearance, meaning he can carry out tasks with little or no supervision, and as any manager knows, the benefits of an agent (“he can act autonomously!”) are also his disadvantages (“oh no, he just acted autonomously!”).
So if you start messing around with it, make sure you’re very, very explicit in your instructions, like “don’t delete anything. I’m serious.” (Thankfully, by default, Claude Code still taps you on the shoulder before something is irreversible.) It’s like raising a 5 year old with superpowers.
Also, keep backups of everything important. But obviously you already do.
Uh, sure… moving on, I understand why this is so important to programmers. But does it really matter for the rest of us?
Of course! As Future Perfect contributing editor Dylan Matthews wrote last year, borrowing a phrase from AI writer and investor Leopold Aschenbrenner, the terrifying endgame is “walk-in remote workers.”
Simply put, if you’re a remote worker, it probably means you perform most of your tasks on a computer. Like I’m doing right now. And although I don’t consider myself a manipulator of computer code in my work, deep down that’s exactly what happens with every letter I press in this document.
Great language models (especially complex reasoning ones like Claude’s Opus 4.5, the preferred model for Claude Code’s supercharged work) are already very good for thinking, analyzing, and writing, and are likely to get better.
Claude Code is what happens when you take a language model and give it access to tools (file editing, searching, command execution) within your codebase, with guardrails that you can loosen (or, unfortunately, remove). In other words, if you are a remote worker, Claude Code could possibly “step in” and do some, most, or maybe all of that work. If chatbots really could advisemodels like Claude Code can actually do.
And Anthropic is already trying to bring that same “Claude with his hands” feeling out of the programmer’s cave and into the rest of your digital life. That’s the idea behind the newly launched Claude Cowork: instead of pointing Claude at a codebase, you point him at a regular person’s folder (his notes, documents, spreadsheets, PDFs, screenshots, the junk drawer of the modern workplace) and he can read, organize, extract and compose within that space to produce real deliverables, not just suggestions.
If Claude Code is a direct remote worker for software teams, Cowork is the version that can be integrated into the work that most remote workers actually do: turning messy input into usable output, faster than you can say “sorry, backing up.”
Yes, now you might understand why Anthropic CEO Dario Amodei warned that we could be “sleepwalking into a white-collar bloodbath” in which AI would quickly wipe out large numbers of entry-level jobs.
Have you tried Claude Cowork?
Not yet. Cowork had only been available on the $100/month Max account and up, though on Friday Anthropic opened it up to $20/month Pro accounts, and I obviously have to save up for the post-work apocalypse.
Wait, aren’t you supposed to be the one? Good news boy?
In fact I am! (Sign up for the newsletter here.) And if you squint, you can argue that what we’re likely to see isn’t so much replacing human jobs as reorganizing them, turning workers into managers of teams of future AI agents, responsible for setting goals, checking results, and making decisions. So I guess in this more optimistic future we will all be office space‘s Bill Lumbergh, leading our army of AI agents to complete endless TPS reports.
O brave world, which has such agents in it!
Yeah, I think the only thing we can count on is that it’s going to get weird. I mean, weirder.
But in the meantime, unless you’re planning on sabotaging data centers (please don’t), you can really significantly improve your work and even your life if you start playing with these tools. The first time you create something that works is a pretty powerful feeling. I can imagine how Mickey felt halfway there. The Sorcerer’s Apprentice.
Do you know… how that ended?
And they all lived happily ever after.
(Disclosure: Future Perfect is funded in part by the BEMC Foundation, whose primary funder was also an early investor in Anthropic; they have no editorial input on our content.)
Update, January 16, 2025, 1:45 pm ET: This story was published on January 16 and has been updated to reflect that Anthropic has opened Cowork to Pro account users.