A lot more tricks can be learned from just watching AI work. Instead of allowing AI to work autonomously, go back to the old days where you manually approve every command the AI runs. Just recently while doing performance optimization work, I found Opus using the `perf` command in ways I didn’t know possible. Just give AI a real task and carefully read what commands are used by the AI to solve the problem; most likely the AI uses a trick or two that you didn’t know existed.
Eh sometimes it overcomplicates things. I have seen Opus 5 do mad shell gymnastics for something that could have been a simple `jq`.
> “the old days”
Is this just plain old rage baiting? I literally can’t tell any more.
It’s a figure of speech. I use that phrase for anything older than three months.
to shreds you say
I find it a valid statement.
IT was always fast, and I explained my students what IDE cable was and they don't even knew about SWAP because they never had memory issues.
This makes me think of something I hate about how LLMs write CLI scripts. Even with the same model on the same day, there's never any consistency with the way CLI arguments are written or handled. Sometimes they want to split on an equal sign, and specifically write that into the code. Other times they write code to pair keys and values based on spacing and order, even and odd.
--date=2026-09-16
--date 20260916
and all variations thereof. It's annoying enough to have to check which script expects what format that I probably should've wasted some time or tokens to align them all, or at least put it into an MD file. But it's just an illustration of how LLMs can make CLI commands more obscure than they need to be.
There are several keyboard shortcuts that you would never find out from watching an agent, but are mindblowing to new Linux users. I suck at remembering vim keybindings, but I have ingrained ctrl+a ctrl+e for jumping to the start/end of a string (which also works all over OS X). I had been a developer for an embarassing amount of time before I discovered those.
There are other terminal specific shortcuts for removing last word (ctrl+w) but they don't seem to be as portable.
The OS-wide Emacs bindings are one of the main things still keeping me on macOS; in editors like Helix (or Zed with Helix mode) I invoke these really cursed combinations of the Helix(vim-like) shortcuts in the editor with the Emacs ones from the OS and it feels really natural as a means of moving around text.
I haven't put in enough effort to replicate the experience on GNOME but once I jump ship from Apple I'll have to.
The way macOS treats shortcuts thanks to the separation of Command and Control is great. It even makes Google Docs pretty painless to use because it's mostly like Emacs. Word still tries to hijack the shortcuts so that's suboptimal.
I think gnome just had this as an option somewhere. Asking chatgpt says that this was removed in GTK4, so maybe it's no longer a thing?
I can still do everything in gnome with a shortcut, but not in vim style.
Another one, not quite as handy but still awesome: ctrl s / ctrl-q. This freezes/unfreezes the terminal output so you can read it, without interrupting your program. Useful for very fast walls of text.
Oh so that is what this is for. I thought ctrl-q / ctrl-s are there as a nasty way to screw with users, who every now and then accidentally press one of these, and find their terminal frozen and no longer visibly reacting to input, for no apparent reason.
That's actually XON/XOFF - in the old days when computers could send serial data to your VT100 or whatever faster than it could handle it (and we're talking 9600 baud here), the terminal would send XOFF (ctrl-S) telling the computer to stop sending - then, when the terminal caught up, it would send XON (ctrl-Q) to let the system know it could resume sending.
The advantage of doing it this way ('software flow control') is it worked with just TX and RX lines, rather than having to connect up additional wires to indicate when the terminal was busy ('hardware flow control').
Ah this is also a fun default feature of Powershell, except it works on mouse click.
there was an old riddle I saw on usenet, "there is a key I use for one of its intended purposes several times a day, but when it stopped working I never missed it. what key was that?". the answer was the caps lock key, one of whose intended purposes was reverting back to normal after accidentally pressing it to go into caps mode.
You can disable this misbegotten feature with
stty -ixon
I remember when the Pause key did that! It's long enough ago that I don't remember which systems did so, and most laptops don't even have a Pause key any more, but it was often very handy for reading fast-moving console output.
It can’t always not interrupt your program. Think about it: after you press Control S to trigger XOFF, the fast output has to be buffered in memory. That memory is certainly not unlimited. There’s no way for the program to keep outputting text. Indeed some programs (including Claude Code, at least a while ago) really don’t like it when you press Control S and let the terminal be frozen for a very long time.
Interesting; I assumed the write to stdout would just hang for the writing program.
ctrl+u and ctrl+k (to delete the line forward/backward from the cursor) are super useful too
ctrl+u generally works for lots of password prompts to reset/remove what was typed, wherever you come across a password prompt and a keyboard including those two keys.
This is interesting, I also do this, it will not let me remember the commands, but, I will learn what is possible and roughly how. And in this day and age that is more important. Just like stamping knowledge into your head also became less important (but not 0 important) with books and the internet.
> go back to the old days where you manually approve every command the AI runs
Those are also the current days if you have any sense. It's a bad idea to run an LLM with access to your machine at all, but if you absolutely must, you better review everything it does to make sure it doesn't run anything insane.
So say people so afraid of LLMs they don't even use them :).
The rest of us picks a level between "approve some" and "YOLO" depending on how much they worry about having to clean up the workspace afterwards. Encouraging using the repo and some dedicated per-session storage, while discouraging changes to global state (like installing packages system-wide) reduces cleanup problem to "every now and then `rm -rf` some agent session dirs".
They run in a container with no network access, no data access other than the project folder.