Agentic systems are probabilistic, the field is young, and the right answer to many important questions is genuinely unknown. Developers who need certainty before they act will find this environment punishing. Developers who can move confidently under uncertainty — making reasonable bets, tracking what they learn, updating as evidence accumulates — will find it energizing.

The discomfort with uncertainty shows up in recognizable patterns. The team that won't deploy until the eval scores are perfect — which means never, because perfect eval scores on probabilistic systems don't exist. The developer who won't commit to a prompt because maybe a different approach would be better — ignoring that the only way to know is to run it. The architect who designs for every possible future requirement, producing a system too complex to reason about in service of flexibility that may never be needed.

Certainty-seeking in an uncertain field doesn't produce certainty. It produces paralysis dressed up as rigor. The eval scores that would justify deployment never arrive because the standard keeps moving. The prompt never gets committed because another alternative always seems worth exploring. The architecture never gets simplified because what if we need that flexibility.

The alternative isn't recklessness. It's calibrated confidence — the ability to assess what you know, what you don't know, and what level of certainty is actually required for the decision at hand. Deploying an agent to handle low-stakes customer inquiries doesn't require certainty about its behavior in every edge case. It requires confidence that the common cases are handled well and the failure modes are recoverable. That's a much more achievable bar, and it's the right bar.

There's also an epistemic honesty argument. The field doesn't have settled answers to many of the important questions. Pretending otherwise — adopting confident positions on things that are genuinely unknown — is a way of performing competence rather than developing it. The developers who are most useful to work with are the ones who can say clearly: here's what I know, here's what I don't, here's my best current bet and why.

Uncertainty is the medium. Learn to work in it.