The most important AI skill leaders can borrow from pilots

Pilots operate under a strict risk management framework that leaders can apply to the deployment of artificial intelligence, according to an opinion piece by Esti Peshin. The author recounts a personal experience of diverting to a third airport due to bad weather, emphasizing that "maybe" is not a risk management strategy. The principle is not to avoid all risk, as flying is inherently risky, but to ensure that any accepted risk has a justifiable return.

Applied to AI, this means questioning not just if a system is risky, but what specific return is expected for that risk. Shipping an unvetted AI model into critical workflows, such as those involving medical records or financial approvals, simply for speed, is deemed an unnecessary risk without commensurate return. Relying on a competitor's speed advantage as the sole justification is characterized as fear disguised as strategy.

The article stresses that risk decisions must be made at the appropriate level. In aviation, the pilot makes the call. Similarly, in AI governance, decisions about deploying models or granting agents access to production systems should be made by the manager who owns the downstream consequences, not by product leaders or engineers under pressure. The key question is: "Who in this workflow has the authority to say no, and does that person have full visibility into what's being deployed?"