Claude Code just got WAY more expensive (called it)

This is why we pay for electricity by usage. Imagine if they said 2K a month unlimited, someone would run a factory on that.

Just a reminder - Theo is an investor at Cursor, Claude Code's competitor.

"Here's the thing about those ai companies. They don't care. They come in, they undercut everything, and they run us out of business, and then, once we're all gone, they jack up the prices" - Michael Scott

If prices keep rising at this rate it'll end up being cheaper just to hire a senior dev

Claude Code was already very popular when Boris and Kat left Anthropic—which is obviously why they were given Head of Eng + Product positions at Cursor. The fact they left Cursor after less than a month after joining clearly tells you all you need to know about the trajectory of the company. Claude Code is light years better experience than Cursor/Windsurf, etc for agentic coding, despite these “limits” which affect small minority of users, who are most likely not working on anything truly productive anyway.

8:06 - This is incorrect. CC is closed source, but you can absolutely use other models by changing the base URL and API key in your environment variables, and use some open source more cheap models instead. It's definitely not designed for it, but it's possible.

Claude cause increased their price. Theo: let me tell you the entire history of ai...

I can't believe that I actually watch the ads on this YouTube channel. I almost *never* watch ads. Great coverage of your sponsors, Theo. I need to be aware of these products so I can understand the current tech space. Just keep your standards high.

I'm amazed at how much people spend on these tools because they don't limit the context of the activity. Just send a few tokens, instead they are sending the entire code base to each question. Edit. I use Copilot and Gemini Code, but I believe this principle applies to all similar tools. ​When you assign a task to the models, you get to select which files to use as context. The key is simply to limit the number of files you send. As a rule of thumb, the quality of the output increases as you limit the context, and the number of tokens you spend plummets. ​Notice that this isn't hard to do if your code's modules are well-organized. To make a simple modification, there's no need to send 15 giant files. ​In fact, I suspect this is how these tools work "under the hood." The difference is that when you send a large amount of code, the tool likely makes multiple API calls to figure out where the code needs to be modified, which consumes an enormous amount of resources. ​For example, take a large production project with 50 files of 300 lines each. The tool might make 50 API calls just to understand what's inside the project. If you send only 3 files instead, the cost will drop drastically.

Wow I'm super early to this video! Thanks for always keeping us updated with software engineering news, you're the source I rely on lol