>> If I give it too many requirements at once, it completely fails

Same experience here, but the trick is knowing what 'too many requirements' consists of. The modern agents (see Google's Antigravity for an example) make a plan beforehand that you review and approve. You can make adjustments and tell it to break the implementation out into incremental phases so things don't go awry.

The typical home surveillance system gives high resolution full coverage of your property but frequently also includes nearby street traffic and the activity at all of the surrounding residences. You can stream the video to a cloud system and potentially share it however you want. Here's one company that ties it all together conveniently for you but there are several others such as Videoloft and Camcloud;
https://www.een.com/

"Capture video and extract AI-driven analytics from virtually any ONVIF-conformant surveillance camera. Get industry-leading features with the cameras you already have."

Unfortunately the cat is out of the bag on this, it is basically unpreventable. Facial recognition can be applied to any video stream or recording from existing surveillance cameras, such as one that's pointed at the entrance to the supermarket where you shop. These rayban glasses are merely a mobile platform for acquiring the video. Packing the recognition tech into such a small package is a demonstration that it can be deployed almost anywhere.

“Face recognition technology on the streets of America poses a uniquely dire threat to the practical anonymity we all rely on,” said Nathan Freed Wessler of the American Civil Liberties Union. “This technology is ripe for abuse.”

And boy, it sure is. A garden variety miscreant will be able to tag and acquire the name, address, phone number, and marital status of all the pretty girls he comes across in the grocery store for future stalking purposes. Potentially vulnerable elderly people can be similarly identified. Wire this into the Flock license plate reader network and the government will have full coverage of your movements, indoors and out.

Yawn, not interested in your squawking.

>> a good primary prompt, and CLAUDE.md files with well defined project definitions, structure, and so on.

It sure does help to have a good idea as to what it is that you want to do and be able to articulate it. The LLM's will give a lot of suggestions if you ask, and many times they are things you haven't thought about. Best to spend the time to get a good implementation plan together with it before you let it do any work.

Anyone would be a fool to think that their identity as an informant would be safe with the US government.

IBM has historically hired most of the workforce right out of college and expected them to stay on indefinitely. Get a little too old and you get laid off. Its cheaper that way. So now they are hiring inexpensive labor to monitor the chatbots.

From the article;
"Since AI tools can handle most routine coding tasks, the company’s junior software developers now spend less time on that and more time working with customers. In the HR department, entry-level staffers now spend time intervening when HR chatbots fall short, correcting output and talking to managers as needed, rather than fielding every question themselves."

I merely asked for some evidence, which I haven't received. I gave some conflicting examples which seem fine to me, and I demonstrated the fallacy of your argument. If you don't like it, I don't care.

"Kiro runs on Anthropic's Claude models but uses Amazon's own tooling"

The models require a front end that feeds them user queries, gives r/w access to local files, etc. and most people will want it to work from within the IDE they normally use. For millions of developers that IDE is Visual Studio Code. There are several companies out there that have forked VS Code to make a custom version that integrates with multiple LLM models. Amazon Kiro is one of them, so for many people this will integrate right in with their work flow; https://kiro.dev/

You can choose from multiple Anthropic models with Kiro but not other brands such as Gemini, GPT, which means less flexibility for the developer. I will say that in my opinion the Claude models are the top performers.
https://kiro.dev/docs/cli/chat...

Here is someone's comparison of Claude Code and Windsurf; https://aiforcode.io/tools/cla...

So I'm not sure what the big fuss is about. Claude Code is "Terminal-native CLI via npm. Native VS Code and JetBrains IDE extensions." Possibly it gives you a larger context window than some other options? Other than that it appears to be much the same.

>> Nothing in your quote stated the equivalence of the time periods past "last for millions of years".

Obviously I didn't claim there was an 'equivalence' over the entire span of the Earth's history. With that argument you could go back in time even farther to claim that the the usual state of the Earth is a primordial cloud of cold dust.

>> Eh. They're right. You will find that evidence yourself if you go look.

According to wikipedia;
"Throughout Earth's climate history (Paleoclimate) its climate has fluctuated between two primary states: greenhouse and icehouse Earth. Both climate states last for millions of years "

So, not usually "tropical all over, with no ice at the poles".

I use a couple of forks of the vscode IDE, Windsurf and google's Antigravity. A basic subscription to Windsurf is $15/month. You can select from a dropdown menu of models, and they cost a varying amount of 'credits' to use. For me the best model is the recently released Claude Opus 4.6, pretty amazing.

>> The usual state for the Earth is tropical all over, with no ice at the poles.

Show evidence.

>> so that the ice caps do not melt for a third time in the last 20 years?

You're still blabbering?

>> like they were supposed to in 2012

No cite of course.

>> They have been CATASTROPHICALLY wrong every single time.

Utter bullshit.