It usually takes more than a year even for employees.

Since git and nodejs are both common in modern development and are expected to work efficiently with huge numbers of files, this was a real bottleneck and it couldn't easily be tackled without threatening backward compatibility.

It encourages one to generalize from instances to the group, and from the group to individuals.

Cherry-pick your examples and you smear the group. Invoke the smeared group, and you target the individual.

It's structurally isomorphic to racism and any other -ism.

Marx is great at building a narrative that generates resentment if you buy his frame. But you don't need to buy his frame, and if you don't, you suffer a lot less resentment. It's no way to live.

A machine that takes longer and longer to prove propositions in ever more inscrutable ways is hardly useful at all.

The machine too needs to produce more generalizable and comprehensible systems, for it to scale up its own conceptualization. Needing to load all the new mathematics in the context window won't be great either.

The Usborne books were the single strongest teacher I had getting into coding. I never owned any; I relied on the library.

In the earliest days, I didn't have a computer either. I'd read an Usborne book, then hang around computer stores poking at the 4+ years out of date Commodore 64s and CPC 464s and even the Acorn Archimedes (fanciest most capable BASIC), putting what I learned into practice. I'd even practice on VTech devices with two monochrome LCD lines of text, in toy shops, to get my fix.

I don't think I'd be where I am today without those books.

Certainly there needs to be repercussions, the kids need to learn, and need to be warned in advance, but kids will do dumb things.

You can talk about news coverage, but you can't force 12 year olds to watch the news, or understand how their actions have consequences. When the consequences do come, the focus needs to be primary on rehabilitation and restitution for the first instance.

In the latest commit, refreshTokenExpiry should be snake case. Fix and make a note to do that in $LANG.

Or otherwise scope the rule. The point is you don't need to be super verbose about it and you get to fix forward too, preventing the same issue reoccurring. You build knowledge and context that let you move faster.

I'm long past the need to code everything by hand; I've written editors, compilers, DOS TSR routines in assembler, disassemblers, debuggers, all sorts. I don't see any coding mountains remaining that I have a burning desire to climb. When I interact with an agent, I'm concerned about architectural nudges and fairly high-level details. And I anticipate climbing further up the abstraction stack over time.

Product management is increasingly vital; it's becoming very, very easy to implement the wrong thing, and you need to rip it out again. Editorial discipline is needed.

There's still room for something vaguely IDE shaped, but it's not going to be code oriented.

Trying to find the stable point of agentic coding is like trying to catch a falling knife. Will you still need to look at diffs? I for one no longer make any edits, as a policy - I either tell the agent to fix it, or tweak a skill or memory or doc so it doesn't make the same mistake a second time, or configure something adversarial. But does that continue indefinitely?

If we create a machine that is able to print on the terminal 'I feel pain', how do we know when to believe the machine is feeling pain?

This isn't enough:

    echo "I feel pain"