Clarke's three laws
Arthur C. Clarke’s three laws started as rules of thumb about prediction. The third one still shows up in product conversations because it names how advanced tooling lands in a room that did not build it.
- When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong.
- The only way of discovering the limits of the possible is to venture a little way past them into the impossible.
- Any sufficiently advanced technology is indistinguishable from magic.
The first law is a warning about authority and conservatism. The second is permission to probe past today’s “possible.” The third is the one that bites in software and operations.
When a system is opaque and fast, observers treat reliability, cost, and failure modes as somebody else’s problem. That is fine for fiction. In production it becomes a demand for outcomes with no appetite for the work that produces them. A slick UI or a confident model reads as the third law in action even when the underlying stack is ordinary databases, queues, and human runbooks.
The useful move is not to sneer at “non-technical” stakeholders. It is to replace magic with inspectable interfaces, the same fight as the law of leaky abstractions when the demo stops working. Contracts, SLOs, budgets, and rollback paths are how you make the trick legible without pretending the stack is simple. If you cannot point to what fails when assumptions break, you are asking people to believe in sorcery. They will. Then they will be angry when the spell stops working.
Clarke himself framed the third law as a limit on imagination, not a license to hand-wave. The same line appears in his essay “Hazards of Prophecy: The Failure of Imagination” in Profiles of the Future and reached a wider technical audience through a short piece in Science in 1968. Treat it as a diagnosis. When your roadmap assumes miracles, you have stopped doing engineering and started writing fantasy with a headcount.