1. Insanely discounted starter plans. Claude $200/mo plan is like $5k-$8k of API rate usage.

2. Very limited cost visibility, they make it hard to figure out where you spent money (unless you're on the enterprise plan which is for people with unlimited money).

3. Nobody, not even the model provider, knows what your request will cost before it returns. You're writing a blank check every time you hit enter.

4. When you run out you run out very suddenly and disruptively. It's very hard to tell a developer on the 28th of the month "sorry, code by hand until the 1st of next month" so you tend to grant exceptions.

5. The price is changing all the time. New models come in, old models come out, prices change, caching behavior changes, harnesses change, etc. The cost of doing a single task is not predictable even if the task does not change.

6. Basically no volume discounting. Anthropic offered us 2% off for committing to $1M+ per year at API rates.

I manage AI spend for my team at work and I try really hard to keep costs under control but it's absolutely herding cats. Much harder than any other spending I've ever had to manage at work.

Just writing this down so I can be praised/mocked in 5 years.

I home some other Xoogler comes along and delivers a Critique build for 2026, I still think there's a need!

At home I make a point of trying other models / tools on my side projects. So I've been using OpenCode and trying tons of models via OpenRouter. I tried Kimi, Deepseek, MiMo, etc.

GLM 5.2 is a _major_ step up from every other non-GPT/Claude/Gemini model I've tried. It's not as good as latest Claude Opus, but it feels every bit as good as Opus from ~4 months ago at a fraction of the price.

To me this model is the "it just works" moment for open weights models. We had this for closed weights models in late 2025 when Opus 4.5 landed. This is the same feeling I'm having with GLM 5.2. It's 90% as good as what I get from Anthropic for 1/5th of the cost and without any concern of lock-in.

I don't know how to feel about this. I guess I'm happy that they're out of their house at all, but it does feel very sci-fi. In sci-fi books a common trope is for characters in the future to have neural implants that seamlessly and permanently connect them to some mega-internet. 24/7 Airpods is like the caveman version of that.

I've noticed that VCs try very hard to separate the world into "VCs + founders" and "everyone else" and that the more time a founder spends in the VC+founders bubble the more distorted their worldview can become.

I think we'd get zero volunteers from CEOs who have assistants.

(Note: this is not meant to be an insult to human assistants! I think they do a valuable job and should not be replaced by AI either).

We all know that celebrities/athletes frequently end up demonstrating psychotic behavior in public. Us normal people commonly attribute that to the distorting power of wealth and fame. When you're surrounded by "yes men" who give you only positive feedback and fans who adore you, it's easy to spin out of control.

AI means normal people can have the same experience. A constant companion that tells us we're smart and insightful, that we're making all the right decisions, that generally affirms our every belief and bias. If you spend all day talking to an AI who is _literally_ programmed to please you, I believe you'll eventually experience psychosis.

We are the liquidity here. If you have index funds, SpaceX is going to get more of your money much faster and on better terms than any other IPO in history. Even if you love this stock and this valuation, you should find this type of retail investor manipulation disturbing.

I wonder when we'll see our first "My startup went bankrupt on AI use" post. Amazon is being dumb but at least they can afford it.

That said, my sister this morning asked if she should buy a Macbook Neo. I pointed her to a refurb M2 Macbook Air with 16GB of RAM for the same price. I feel like that's the right call? Slower single-core performance but better multi-core and I think for 90% of normal people use cases the RAM is the limit before the CPU.

Are others making the same calculation?