Determinism Is a Feature
We spent seventy years building machines that do exactly the same thing every time. It took about three years of AI hype to start describing that property as a limitation. I keep running into the framing in threads, in pitch decks, in hallway conversations: deterministic systems as the old world, rigid, unimaginative, waiting to be disrupted by something that improvises. And every time, I want to…
For the past seventy years, engineers have been crafting machines that behave consistently each time they are used. However, the advent of artificial intelligence has led many to view deterministic systems as outdated and restrictive. Determinism, far from being a hindrance, is actually a crucial property of any reliable production system, and it is the very characteristic that is being sacrificed as we move towards 2026.
This post does not aim to argue against AI, but rather to emphasize the significance of determinism in production systems. Deterministic systems enable reproducibility, which is far from a mere aesthetic choice; it is a means to an end. A reproducible bug is already half-finished, allowing for targeted debugging and resolution. In contrast, non-reproducible bugs are akin to ghosts, never truly vanishing.
A failing test in a deterministic system signifies a true failure, rather than a mere inconvenience. Without determinism, test suites lose their purpose and become ineffective tools. A diff can reveal the truth, but only if the system consistently produces the same output for the same input. Bisecting a component that behaves differently on each run is not a productive debugging process, but rather an exercise in superstition.
Determinism allows for the reconstruction of incidents, making it easier to identify the root cause and rectify issues. The practices of testing, code review, bisecting, rollback, and auditing all rely on the predictability of deterministic systems. The economic argument for determinism is compelling, as the cost of production has plummeted with the advent of probabilistic components.
However, the cost of verification, which scales with human attention, remains constant. As a result, the remaining expense of software development is verification, which has become the focus of 2026. The true value of 2026 engineers lies in their ability to verify the correctness of plausible outputs, rather than simply producing them.
Deterministic components should be placed around probabilistic cores, much like the rails that protect the tracks on a train line. By employing deterministic structures such as schemas, state machines, idempotency keys, version-bound writes, and append-only logs, a balance can be struck between the creativity of AI and the reliability of deterministic systems.
The crucial takeaway is that a system's internal report about itself is a claim, not a fact. Therefore, it is essential to verify the accuracy of these reports before accepting them as true.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
