Augment Code Blog

> Nothing on infra costs, hardware throughput + capacity (accounting for hidden tokens) & depreciation

That's because it's coming at things from the other end: since we can't be sure exactly what companies are doing, we're just going to look at the actual market incentives and pricing available and try to work backwards from there. And to be fair, it also cites, for instance, deepseek's paper where they talk about what their power foot margins are on inference.

> just a blind faith that pricing by providers "covers all costs and more".

It's not blind faith. I think they make a really good argument for why the pricing by providers almost certainly does cover all the costs and more. Again, including citing white papers by some of those providers.

> Naive estimate of 1000 tokens per search using some simplistic queries, exactly the kind of usage you don't need or want an LLM for.

Those token estimates were for comparing to search pricing to establish whether — relative to other things on the market — LLMs were expensive, so obviously they wanted to choose something where the domain is similar to search. That wasn't for determining whether inference was profitable or not in itself, and has absolutely no bearing on that.

> Doesn't account at all for chain-of-thought (hidden tokens) that count as output tokens by the providers but are not present in the output (surprise).

Most open-source providers provide thinking tokens in the output. Just separated by some tokens so…

Nice, only got the pro plan and didn't want to get Nickel and Dimed per usage so ended up ignoring claude code, but will give it another go now to see how far I can get with the Pro plan.

Turned out to be a blessing as it led me to discovering augmentcode.com which I'm now using. It's a monthly usage cap but has a unique context engine where it indexes your entire code base which gives its agents access to a lot more content about your code base which helps when adding features to match your existing implementations without needing to explicitly feed it context in every agent message. It still uses Claude Sonnet 4 [1] under the hood so the results are great.

Blows me away what AI agents are capable of these days, it's especially nice when they end up creating new Tailwind UIs (Vue) that's better looking than what I was even thinking about creating.

[1] https://www.augmentcode.com/blog/claude-sonnet-4-the-best-model-with-the-best-context-engine

I wonder if Augment [1] are working on a Zed plugin.

I've been using Augment for more than a year in Jetbrains IDEs, and been very impressed by it, both the autocomplete and the Cursor-style agent. I've looked at Cursor and couldn't figure out why anyone needed to use a dedicated IDE when Augment exists as a plugin. Colleagues who have used Cursor have switched to Augment and say it's better.

Seems to me like Augment is an AI tool flying under most people's radar; not sure why it's not all over Hacker News.

[1] https://www.augmentcode.com/