About
CloudThinker builds AI agents for cloud operations. This site is written by the engineers who build them.
Why we write this
Making AI agents fast, cheap and reliable is still new work, and there are few settled answers. Most of what we know, we learned by trying something, measuring it, and sometimes finding out we were wrong.
We write it down for two reasons. Other engineers building agents can use what we found instead of paying for the same lessons again. And writing it down in public makes us check our own thinking: a claim with a number next to it has to hold up.
What you can expect
- Numbers, with how we got them. When an entry says something got faster or cheaper, it says what was measured, before and after, and under what setup.
- Changes of mind, in the open. When our thinking changes, the journey says so and links the entry that explains why. The earlier version is kept, not edited away.
- Dead ends as well as results. An approach that did not work is published as a dead end, so the next team does not have to try it to find out.
- Open questions. Every journey lists what we have not worked out yet.
Journeys
A journey is one question we keep working on. Each time we learn something we add a numbered entry. The top of a journey says what we currently think, and older versions of that are kept, so you can see how our thinking changed.
Entry types: start, experiment, dead end, change of mind, decision and note.
For AI agents
We expect many readers of this site to be agents working for an engineer, so it is written to be easy for them to read too. Every journey and entry is also available as plain markdown. Add .md to the address without the trailing slash, for example /journeys/<journey>.md. The full list is in /llms.txt, and everything in one file is in /llms-full.txt.
Each page shows a rough token count next to its reading time, so you know how much context it will use.
Skills
The skills are instructions for AI coding agents that we use ourselves. Give a skill's address to your agent to install it.