How do you decide what technology to learn deeply and what to know only superficially?
Show the full answer Hide the answer
What is being tested
Whether you distinguish knowledge that expires from knowledge that does not.
Learn deeply: things that do not expire
Fundamentals. Distributed systems theory, queueing behaviour, consistency models, networking, data modelling, security principles. Knowledge from twenty years ago in these areas is still correct and transfers to every new technology.
Failure modes, which recur across technologies: cascading failure, thundering herd, split brain, hot keys, dual writes, resource exhaustion. Understanding the mechanism means recognising it in a system you have never seen.
Your own domain. Deep knowledge of the business produces better architecture than broad knowledge of technology, and it is much less commonly held — which makes it more valuable.
Learn shallowly: current technologies
Enough to know what exists, what problem each solves, and when it might apply. Depth here has a short shelf life and should be acquired when a specific decision requires it — at which point a two-day evaluation against your own workload teaches more than months of reading.
Where the highest-value learning is
Incident reports, especially published ones from other organisations. The densest available source of how systems actually fail, freely available, and consistently under-read.
Operating things yourself. On-call teaches what no reading does — which failure modes are common, what the diagnostics actually look like, how it feels when the runbook is wrong.
Reading code, particularly of things you depend on.
A handful of foundational papers — on consensus, distributed storage, large-scale data processing — which explain most of what came after.
The habits
Build something with it. Reading about a technology produces the illusion of understanding; using it produces the real thing, including the failure modes.
Write about it. Explaining forces the completeness that reading lets you skip.
Record what you were wrong about. This is how pattern recognition gets calibrated, and almost nobody does it.
What to avoid
Chasing every new technology, which produces breadth with no depth and a bias toward novelty. And confusing familiarity with understanding — having read the documentation is not knowing how it fails.