Same here. Codex support is a recent addition however, and it’s not clear if MCP servers and other rules apply to Codex. Also it would be nice to be able to just have a session working on the main branch as concurrent work in worktrees can get messy

No it's an instruction for humans

Funny, that didn't occur to me at all. :-/ Maybe that's because I'm used to the FHS and I expected a path starting with /System to be either a path that is interpreted by some command or a description of an UI flow, not a path in the file system. So the thing you would type in the shell is just '/System/Applications/Utilities/System Information.app'? Does the Finder support starting programs by typing the path in the pathbar, like MS Explorer on Windows?

Intuitively feels like this would lead to less empathy on average. Could be wrong though.

> casuals who want to spend as little money as possible to use lots of datacenter time as their artificial buddy

What kind of elitist bs is that?

Crazy to read this on two hours sleep in two days because of startup deadlines

Good point, OP describes "firms" e.g. large established organizations, and ignores early stage startups for the most part. I think that fact and your comment illuminate another reason for the high salaries - namely:

A huge raison detre to create and sustain a "firm" in the first place is stability. That stability extends on all fronts - stable base of customers, stable base of revenue, stable base of shareholder returns, and stable base of compensation for employees. Therefore, it is expected that there would be some looseness/slip/"irrationally high" or however you want to think about it, in the salaries. An early-stage startup by comparison is dealing with a different problem and purpose.

This might be kind of tangential, but—if people are working pretty hard at early stage start-ups with the intent of having their start-up bought (presumably by a large company), then the large companies must be doing something, right? Clearly the startup employees think it is worthwhile to create a good product (the company to be bought). Which means they think there will be some reward based on the quality of the start-up they create. Which means they think the large company has some way of distinguishing between high and low quality start-ups.

That’s a thing the corporation does, I guess.

And stable labor availability for high load periods, even if rare.

Seriously, my last two roles at startups have been the most work of my career (including earlier work at startups). Everyone I know is pretty frequently working nights and weekends to make sure everything gets shipped. The “mindless code” is mostly written by AI so, at least for me, most of the code I’m writing requires a white board to work out that the math is correct. At the other end of my current company people are physically managing warehouses; certainly looks like work to me.

I’ve had cushy jobs that the author is mistaking for “all jobs”, but my experience is that these places are in decline and when layoffs come, get ready to work hard if you want another job.

Once you pass 1000 engineering headcount, this starts to happen.

Especially once the org is old enough that the 2-year tenure people start to be in the 50th percentile of seniority.

Crazy to read this managing 6 soon to be 7 projects at a startup but actually just working like 20 hours a week

Is your startup growing? And at what rate?

We have 1 big customer with a yearly license of our product and some sprinkled around without license. Not growing quickly thankfully.

I checked and my average over the last 3 months is 5hr16min *(cry emoji)

You should be getting a full 8. Dont neglect your health at this age. Youth is one thing you cannot get back.

I'm trying desperately, trust. I also have central sleep apnea, its very hard.

Couldn’t you just pipe the stdout to another dumbpipe

Not a very friendly API

This works:

Remote:

  $ socat TCP-LISTEN:4321,reuseaddr,fork EXEC:"bash -li",pty,stderr,setsid,sigint,rawer&
  $ dumbpipe listen-tcp --host 127.0.0.1:4321
  using secret key fe82...7efd
  Forwarding incoming requests to '127.0.0.1:4321'.
  To connect, use e.g.:
  dumbpipe connect-tcp nodeabj...wkqay

Local:

  $ dumbpipe connect-tcp --addr 127.0.0.1:4321 nodeabj...wkqay&
  using secret key fe82...7efd
  $ nc 127.0.0.1 4321
  root@localhost:~#

Claude runs on AWS afaik. And OAI on Azure. Edit: oh okay maybe GCP too then. I’m personally having no problem using Claude Code though.

Right. Look at Electron apps. They're ubiquitous despite the poorer performance and user experience because the benefits outweigh the negatives.

Maintaining a codebase isn't going to be a thing in the future, at least not in the traditional/current sense.

Seems like a use case for language models.

No one expected that in the future we would be reading AI generated poetry to cats in strollers.