People keep asking me what they should read to get up to speed on enterprise AI. The honest answer is that most of the useful material is not in books yet. It is in incident reports, post-mortems, and the conversations that happen after things go wrong.
But there is a shortlist of frameworks that I find myself returning to consistently, and since I have been asked this enough times that it feels worth writing down properly, here it is.
The first thing I send is not about AI at all. It is Eugene Wei's essay "Invisible Asymptotes." The concept, that every product and every business has a hidden ceiling that is not visible until you are close to it, applies directly to how organizations should think about AI adoption. The productivity gains from AI tools have invisible asymptotes. Knowing that they exist and thinking about where yours might be is more useful than assuming the growth compounds indefinitely.
The second is a paper rather than a popular article: "Taxonomy of Failure Modes in LLM-Based Systems" from a team at DeepMind. It is dense and technical in places but the failure taxonomy in the first half is exactly what a COO needs to understand before making infrastructure decisions. The failure modes of AI systems are categorically different from the failure modes of traditional software and the people making decisions about these systems need to understand that.
The third is something I come back to quarterly: the AI incident database maintained by the Partnership on AI. Real incidents, real organizations, real consequences. Reading through ten of these before any major AI deployment decision resets expectations in a way that vendor demos consistently fail to do.
The fourth is less about AI specifically and more about managing technology adoption in complex organizations. "Diffusion of Innovations" by Everett Rogers is from 1962 and it describes the adoption curve dynamics that play out in AI deployments almost exactly. The innovators, early adopters, early majority distinction is not just a marketing concept. It is a useful operational framework for sequencing rollout decisions.
The fifth I hesitate to include because it sounds self-serving but it is genuinely the most practically useful category: talk to operators at companies 18 to 24 months ahead of your own AI maturity. Not vendors, not consultants, not researchers. People who deployed what you are deploying, encountered the problems you have not encountered yet, and made the mistakes that are still ahead of you. The information density in one hour of that conversation exceeds almost anything written down.
The reading list is not about becoming an AI expert. It is about building the mental models that let you ask better questions and make better decisions when the AI vendors, the internal champions, and the board are all pulling you in different directions with different levels of information and different motivations.
The decisions you make in the next eighteen months will be consequential for a long time. The preparation that pays off is not knowing the most about the technology. It is knowing enough to recognize when you are being told something that does not hold up.
Top comments (0)