Comment by splwjs
5 days ago
There's already an obvious stench to "you should scale down your engineering team to a skeleton crew whose core competency is using our product, so that it's the only way to modify your product"; that's going to result in a lot of foodless tables when anthropic et al decide they have enough leverage to stop subsidizing their subscription prices down to what, 4-10% of the marginal cost? Well it doesn't matter how much they want to jack the prices up, if your engineering team requires tokens to do anything you'll just shut up and pay whatever it is.
There's another big problem with the blackbox shrugoff of "no, there's no way to know how many tokens a given request will cost, idk just assign an agent to that or something lol"
But now the software may just decide for itself that your application of it needs to be silently diverted onto a snipe hunting trail. Surely they'll only ever do this for anyone developing a competing product. Or malware. Or Criminal activity. Or one of ten other applications that the system will never misjudge.
You don't need a datacenter the size of Ohio to figure out that agentic ai maximalism is going to hurt you more than help you.
Repeat of the cycle where companies moved to the cloud, now locked in.
Move engineering to Claude, then locked in.
What’s played out at the infra level will now play out at the software engineer level… is that analogous…?
Spot on, the size of backend teams reduced drastically as we made the progression through cloud, serverless, SaaS, iPaaS, and now agent orchestration tooling.
What would be a team of about 20 devs a decade ago, are usually about 5 persons across all project roles.
Hence when someone says their job is AI safe, I can only understand they weren't affected by such "progress".
I'd argue it's different because ephemeral, vps, baremetal, self-hosted, the skillset is still "managing a linux server" and the most radical shift is to an expectation that a server is a thing that is created and destroyed on demand (so now you need to script deployment), which isn't really a radical shift away from the skillset.
The goal of AI seems to pretty explicitly be "stop coding; from now on the mechanic fixes your car", which I would argue is a very different shift.
Also if you host a criminal content on AWS they probably close your account and ban you instead of silently rerouting all traffic away from your server (which may just be hosting a game where you kill goblins or steal cars or something) and refusing to acknowledge that that's what's happening.
Ya if you just run VMs but now a lot of cloud vendors have you build stuff using their proprietary features so decoupling becomes harder and can be risky.
Remember that top tier senior dev you hired? Now you never know when they're going to automatically and silently fall back to junior level. And randomly back again.
That may be the equivalent to where this is heading
Given the cost overruns and rework from AI swarms... I feel that the AI Gatekeeper model is a much better approach to how to leverage AI for most things... One person, directing one agent and babysitting to understand, correct and gate the results.
These AI companies have never been incentivized to make your code better, but I guess they're saying the quiet part out loud now.