Mostly Fine

Where so many books about AI offer either evangelism or panic, Szalma offers operations, governance, and a reality check.
Every era of technological mania eventually produces a book that refuses to join the choir. Mostly Fine is that book for the AI age: funny, fiercely pragmatic, and written from inside the machine room rather than from the TED stage.

Szalma has managed people, built data infrastructure, and built corporate AI systems himself. His central insight, that businesses run on tacit knowledge while AI models are trained mostly on polished conclusions, is one of the more illuminating frameworks yet offered for why AI looks dazzling in demos and mediocre in production.

Szalma methodically dissects the limitations of large language models, maps the enterprise failure modes, the security risks, and the organizational blind spots that follow from them, and offers concrete remedies, all in roughly 140 pages of mercilessly edited prose. It is the kind of book you finish on a single flight and reference for the rest of the year.

Mostly Fine gives readers something rare in AI publishing: a path to ROI.