The pattern is pervasive. Big corp promotes a solution that fits their need. People read about it, think adopting big corp solution means they are doing the right thing. Few people have big corp need, let alone everyone big corps are different. And then endless hours spent fighting big corp solution to not so big corp problem.

You can't force people to feel what you feel. One can (pr|t)each, without experiencing, other can only mimic or rebel. That's how cult is formed.

I had the same suspicion so made this to examine where my tokens went.

Claude code caches a big chunk of context (all messages of current session). While a lot of data is going through network, in ccaudit itself, 98% is context is from cache.

Granted, to view the actual system prompt used by claude, one can only inspect network request. Otherwise best guess is token use in first exchange with Claude.

https://github.com/kmcheung12/ccaudit

I got kinda obsessed with observability a month ago and wired together a full stack for personal use.

https://github.com/simple10/agent-super-spy - llm proxy + http MiTM proxy + LLMetry + other goodies

https://github.com/simple10/agents-observe - fancier claude hooks dashboard

It started as a need to keep an eye on OpenClaw but is incredibly useful for really understanding any agent harness at the raw LLM request level.

A browser extension to add a table of content widget into the chatbot pages (claude/Grok/Chatgpt). Making long conversation easier to navigate. Mainly used on firefox. Not tested on chrome

https://github.com/kmcheung12/tocic

So simple, but brilliant! Il check this out today.

Do you plan on submitting it to the store?

It's so simple that I didn't think much of it. Maybe I should

The analogy I used with the team was that, set the goal, present the map, and figure how to make a better map. Drucker was about the goal with a given map. It is not uncommon for people receiving the OKR not resonating with it. Sometimes they actually have insight into making a better map, but if OKR is OKR, one just have to follow, people swallow their thoughts

Did not expect to see a book about night sky in this post. The book looks great. I'm getting one for my sons.

Hope they like it! Let me know what they think!

The had a currency swap. US earns interest from argentina by taking USD$20B worth of pesos. Argentina get USD$20B. By the end of the swap, Argentina "swap" the borrowed USD$20B with their pesos. Argentina has to pay interest (quick googling didn't reveal the exact terms). They both get their own currency in the end.

> They both get their own currency in the end.

Unless Argentina defaults, as they have many times before.

Exactly. I know I’m not learning much. But it is a healthy/less harmful distraction and I managed to read some Spanish tweet. Can’t complain.

I find your article valuable. It shows me what amount of configuration is needed for a reasonable expectation of performance. In real world, I’m not going to spend effort maxing out configuring a single piece of tool. Not being the most performing config on either of the tools is the least of my concern. Picking either of them, or as you suggested, Postgres, and then worry about getting one billion requests to the service is far more important

I was using the same setup to try to transcribe a sound track of a video. A 60s aac audio took me maybe 10 minutes. I'm on a apple M4 and ran `whisper audio.aac --model medium --fp16 False --language Japanese`. Wonder if I'm doing something wrong