I use AI assisted coding in Agent mode daily. I see it as something similar to AutoTune for vocals - it'll make a good singer great, and a great singer world class. Anything less than a good singer, autotune is going to move to the wrong note and make it worse.

It's the same with coding - it'll make a good and great developers better, and mediocre developers worse. It's probably also important to acknowledge how I'm using it as it's likely that your experience will vary depending on environmental variables-

-I use Opus and Sonnet 4.5 through github copilot in VS2026 in c#/.net 8.0+ projects. This works astoundingly well, particularly Opus in complex projects.
-Gemini 2.5 and OpenAI GPT 5 were almost completely useless
-Gemini 3 was better, but is still behind Sonnet 4.5, and Opus is a whole other league ahead of Sonnet in complex tasks.
-Grok is terrifying. I sent it a prompt once and it YOLO'd its way into completely nonfunctional code while completely misunderstanding the task.

I've never lacked for ideas, just the time to implement those ideas. Anthropic's models have greatly improved my productivity to the point where I can almost keep up with my ideas.

This is factually incorrect. Gelsinger was CEO from February of 2021 until December of 2024. The misses that you mention happened under the previous CEO and Gelsinger can be credited with Intel Arc and Battlemage, 14th gen processors, and the 2nm fab processes that are now coming online.

Given the development timelines of these technologies I expect we'll see a couple more hits that originated in the Gelsinger era before the company will start to struggle again due to lack of engineering minded leadership.

Tell me you don't know what you're talking about without actually saying it.

That 128gb of RAM is assignable up to 96gb of VRAM in windows and I've heard rumors of up to 110gb in linux. There are 3 products I'm aware of that are comparable-

Apple Mac Studio 128gb ram - $4800
Nvidia Digits - $3000 MSRP, not yet shipping as far as I know
HP Z2 Mini G1a - "Coming soon", no pricing info available

It's fine if it's not for you, but you're simply wrong if you think your 6 year old laptop or any current traditional desktop can do what this is built for.

As far as the 12" laptop being a toy, I think that's kind of the point. This laptop is perfect for a child's laptop or student edition computer for younger students. There are a ton of compromises relative to a high end laptop, but I think I would describe it as "stupidly fast for its intended purpose". We will see on the price, but if they manage to come in at or under $500 it's a very compelling offering.

That's the thing though - you can't find what this desktop does any cheaper. The top end configuration with 128gb of RAM is targeted at a specific market segment where the closest two options are an Apple Mac Studio at $4800, and an NVidia Digits with MSRP of $3000 (so $5000+?). It's ok for it to not be made for you, and it's OK for you to find more value from cheaper/different configured desktops, but don't dismiss it for what it is, which is a very compelling offering in the AI/Local LLM space.

The soldered RAM is an AMD thing and it's required for this chip to perform to design specs. There's an LTT video that dropped shortly after the reveal where the CEO goes into more depth where they pushed AMD really hard on the issue and AMD assigned an engineer to do a deep dive on it, only to find out that it's not possible to make the ram replaceable and maintain performance. As for who it's for? Me, it's for me, and apparently a TON of other people agree as I placed my preorder fairly quickly after it went live and still ended up in Batch 2. I'm both a casual gamer and a developer who is stepping into local LLM development. The idea that you can run a 70b LLM comfortably for under $2500 all in, and there's a path to running the full 600+b models without spending nearly $100k is a game changer.

This may depend on the jurisdiction, but I believe there's precedence for "anything bad that happens during/as a result of the commission of a crime is the responsibility of the person committing the crime."

I'm surprised nobody has mentioned this. Printed manuals are out of date before they are off the printer. It's much easier to update a PDF online and just provide links to the always up to date PDF.

Contrary to popular belief on this site, most business code is closed source. However, I am a solid middle of the pack developer and if I can do it just from the Avatax API docs, I'm sure you can too.

This is false. Do this at your own risk.

This is wrong. Well, not all wrong, but the only correct part of this is that it is an easy technology fix. That technology is an API call to a tax calculation service (there are many of these).

Zipcode/Taxrate table is woefully oversimplified to the problem at hand. Zip codes cross county/city lines frequently, which each have their own general sales tax rates. Additionally, there are tax rates by product classification. Prepared food will be taxed differently than unprepared, a diamond ring will be taxed as a luxury good, there are tax holidays where anything classified a "school supply" has no sales tax, but only for a certain day, or set of days. Do not write your own sales tax calculations. There are services available to calculate it for free, and pay options for services that will file taxes for you.

This is a solved problem. I wrote an integration with Avatax (one of many competitors in this field), in about 4 hours, fully tested.

Apparently you've never been in Arizona in late July/August.

PHP web hosting - http://www.phpwebhosting.com/. Once had a 2mb graphic file hosted there that got linked to from the World of Warcraft main page. By the time all was said and done our logs showed multiple terabytes of data transfer over the course of a few days and their servers never flinched.