About
The same sword, in two different hands
AI is a sword. In a ninja’s hands it is decisive; in an amateur’s it is mostly a danger to the person holding it. Same blade, opposite outcomes — and the difference is never the sword. It is what the hand already knows.
Why it exists
AI has made it cheap to produce a configuration, a query plan, a retry policy, a Kubernetes manifest. It has not made it any cheaper to know whether the thing produced is right. That judgment comes from understanding the mechanism underneath — what the shuffle actually does with a skewed key, why VACUUM does not return disk, what a consumer group does during the rebalance it was just told to avoid.
An engineer with that understanding gets leverage from AI: they ask for the right thing, recognise the wrong answer, and catch the condition the model left out. An engineer without it gets fluent output they cannot evaluate, and finds out in production which kind it was. AI amplifies whatever you bring to it. This site is about making sure what you bring is deep.
Mechanisms, not metaphors
The concepts that cost the most in production are not the ones people find hard. They are the ones people believe they already understand. Prose is bad at fixing that, because you can nod along to a paragraph without ever testing what you believe. A simulation is not so forgiving: you change the thing you think matters and the number does not move. That is the moment every lesson here is built around.
So every lesson is a working model of the real thing — move a parameter and watch the mechanism react, including into the states where the usual advice stops working. Each one names the specific misconception it exists to delete.
How it is built
- Every claim is checked against a primary source. The project’s source code, its RFCs, its papers — never a blog post. Blog posts are useful for finding what people get wrong; they do not establish what is true.
- Every number in the prose is produced by the simulation on the page. Each lesson is replayed instruction by instruction and every quoted figure is asserted against the readout the reader actually sees at that point.
- Where the documentation and the source disagree, the lesson says so. That gap is usually the most valuable thing on the page.
- Lessons get pulled when they are wrong. A lesson whose simulation contradicts its own argument comes down until it is fixed, rather than staying up because it is already published.
Free, all of it
All 101 lessons across 12 systems are free to read. No account, no paywall, no email wall. If something here saves you an afternoon, that is the whole point.
Who made it
Kamal Kumar. Corrections and arguments are welcome — especially arguments. If a lesson is wrong, I would rather know.