This is the discipline's university textbook: agents and environments, search, constraint satisfaction, logic, planning, reasoning under uncertainty, decision theory, learning, deep networks, language, robotics, and a closing section on consequences. It is consulted, not read; anyone claiming to have read it cover to cover in one go probably did not.
What I did with it
I recovered the definition of an agent and use it to organise conversations. An agent perceives an environment, acts on it, and is judged by a performance measure defined in advance. If a proposal does not state what it perceives, what it may do, and how you measure whether it did well, there is no agent: there is generated text. That single requirement changes the conversation with a buyer.
The second thing I use sits in the search and constraint chapters. Assigning visits, building shifts, loading trucks, fitting schedules: all of that has a classical, exact and explainable solution, and today it is being handed to language models because that is the fashion. The result costs more, runs slower and cannot be audited. Knowing the other tool exists is what lets you offer the cheap one.
And the third is decision theory: a system deciding under uncertainty needs someone to say what each kind of error costs. That is not a technical question, it belongs to the business, and it is almost never answered.
What I left out
Most of it, without remorse. It is a textbook with exercises and proofs, built for a semester course. The deep learning chapters were already slightly short against the pace of the field, and there is fresher material for that. What does not age are the foundations and the vocabulary, which is why it stays within reach on the shelf.