Listen, I’m a VERY healthy skeptic of AI on a number of fronts, but saying it can’t speed things up is just not true.
I feel like I understand where you’re trying to come from with the argument about skill degradation, but the flat statement you say here is undeniably false.
I finally found it, it’s like 300 pages of individual rejections you need to go. Hard pass.
Not listing the specific game release, more the game epoch:
You world be very surprised. It’s not what I directly work on, but there are entire programs dedicated to tracking and lowering the cost of password and account resets (and doing it securely). That last part is the hard thing. Large organizations can save millions of dollars if they can get people to not lose/forget their credentials -or- self service those resets (but even then, having that person do it is now losing their time and energy).
It’s not just indirect opportunity costs either, it’s direct costs to the IT organization. Smaller teams can absorb this stuff as a rounding error on their time, but the second you’re counting in the thousands, you need to deal with this type of thing. (And I’ll repeat: securely)
Now imagine doing it as a B2C+B2B organization like Google or Microsoft. The costs are staggering.
Not the commenter, but they’re not taking about CPU costs.
I prefer the term “Meat Proxy”, but I get that it’s more of an ad hominem. I use both terms.
This guy stole my brain and dumped out all the thoughts I have.
I will maintain that we have flying cars, and we have had them for some time. They’re called helicopters.
Cursor’s biggest advantage is that it has all the code context loaded right up there in the embedded VSCode editor fork. Augment Code has a similar “context engine” built up, but it’s not a full blown editor.
Cursor’s latest plans were bringing it into something closer to what Claude Code or Claude Desktop is doing (Codex has similar tools.) I’m personally ditching it all and doing VSCodium in 1 window and OpenCode on a terminal. After using Cursor for a year (personal projects, not at work), I can say that it was good, but Composer 1 and 2 were absolute shit and I relied on GPT and Claude models. OpenCode + OpenCode Go (GLM, etc) offer a much more affordable alternative.
I bounced around a bit after Storj decided to hate consumers and institute massive minimums.
I’ve landed on Backblaze B2, S3 compatible. Add others have mentioned, doing a shared NAS 2 NAS hookup with a friend or family member is a nice ideal way to do it without monthly fees.
I use Borg backup from my PCs to a NAS (QNAP), and from there using the built in backup software to go off-site. My self hosted PCs will mount NAS drives directly (like immich and audiobookshelf) and those are also in the NAS off-site job.
Reolink stores to a SD card with the option to upload to FTP backups.
I suspect a big 128 gb card would be good enough without off-site storage.
I’ve run this experiment on myself a great number of times.
Your story is correct, sometimes. It’s not correct all of the time. Sometimes AI is slower and the output is better (think: pair programming). Sometimes AI gets me the first draft and the cognitive understanding takes time (overall, for me, usually faster, but sometimes slower I will readily admit). I’m really good at reading code, I guess? And still other times it will one shot near perfection (in true slot machine fashion). It’ll also produce absolute shit, as well, frequently.
My firm point to you is it’s not a blanket truth.