Let's get real, the AI didn't break into these systems all by itself. It did what it was prompted to do. Now sure, maybe the person doing the prompting didn't realize the implications of what he was prompting it to do. But no, the AI is not conscious or self-controlling, as much as Anthropic would like you to think so.

How about a button that marks an unsolicited connection requests as spam. Garbage connection requests are worse than people posting AI slop, because I never look at the "feed".

Thanks for clarifying, I missed the sarcasm.

Actually, I don't think we were misinformed.

The reason so many companies spend wildly on Azure or AWS, is that they don't properly manage their resources. When you use the cloud, you have to think more about what you actually need.

The problem is a habit that came from the days when everybody self-hosted. They would buy a machine, stand it up, and use it. When that machine wasn't as useful, it wasn't costing them anything to keep it sitting there, so they just let it sit there, idle, or under-utilized. They *were* actually paying for this unused infrastructure, but it was "sunk cost" so they moved on and ignored it.

In the cloud, you have to keep paying monthly for every instance of everything that has been provisioned, (typically) even if you're not using it. Many companies neglect to take down instances they're no longer using. My company found that they could save $20K per month just be cleaning up old, unused instances of servers.

So if a company is disorganized and doesn't have the bandwidth to clean up old, unused infrastructure, then yes, the cloud will cost them dearly. For those that are disciplined, the cloud *can* actually save a lot of money compared to purchased hardware.

If you think "managing a team of AI coding agents" is an isomorphism for Brooks' team structure, you should re-read the book.

His structure was very explicit, and had only exactly two programmers on a team: the "Chief programmer (surgeon)" and a "Copilot".
All the other roles were auxiliary
- Administrator (scheduling),
- Editor (documentation),
- Toolsmith (creating tools to enable the programming process),
- Tester (QA),
- "Language Lawyer" (kind of an architect).

You can suggest that a programmer with a "team" of AI coding agents is a think that works, but don't point to the Mythical Man Month, it has no relation to the pattern devised by Brooks.

I'm willing to bet it will be easier to apply controls to LLMs, than to people.

If your company does phishing tests, and it should, they'll find that 20-30% of people will fall for it. We might have "developed administrative and technical controls" but it's only marginally successful, at best.

The end user always pays for things they use.

Renting is a perfectly legitimate way to obtain the use of something, whether it's a house, a car, or a server. Sure, you might be a higher rate per month than if you buy, but you also don't have to pay to maintain the thing you're renting. That's the owner's job.

Renting is a tradeoff. Buying is a tradeoff. As a homeowner, I can tell you that buying isn't for the faint of heart. You don't pay monthly rent, but you DO pay big air conditioner replacement costs. Hardware infrastructure is no different. You pay for it one way or the other.

If you can convince a person to mistake you for someone who has the right or authority to receive sensitive information, they'll give it to you. This is literally how phishing works.

Absolutely! Robots have an off switch. That would be better.

Every time you experience something, every time you learn something, every time you see something, or do something, your brain gets re-wired. That's literally how the brain works.

So a better headline would be, "Your brain can learn to truly multitask."

This is also kind of obvious, if you think about it. Right now I'm typing, and reading as I type to make sure that what I typed is what I wanted to type. We all eat and carry on conversations with friends at the same time. We all drive and mess with our phones at the same time. Wait, what? I wasn't supposed to admit that!

In this instance, you and I agree completely.

It was a "rogue AI agent" of course!

"Rogue" AI my foot.

Back when humans wrote software, I had a coworker who, when a bug was found in his code, he would always *assume* it was a "Microsoft bug." But on further analysis, it always turned out to be just a plain old, ordinary bug that he himself introduced.

This is the AI version of that.

If we knew the full truth, *somebody* prompted that AI to go look for credentials and try to use them to access those sites. The AI didn't "decide" to do this all on its own.

Not in my experience. I've tried Claude Sonnet and Opus, Gemini 3.1, 3.5, 3.6, GPT Codex, GPT 5.5. I've not tried Grok, though I have doubts about even going there. They all pull the same stunts.

The Mythical Man-Month is a classic. I've read it twice, and given talks on it.

How does it fit into the AI discussion?