Textbooks in Tokenland - Systems Approach

Skip to content

Artificial Intelligence, Teaching, Writing

By

Bruce Davie

June 1, 2026

Textbooks in Tokenland

It has been a few years since a colleague first suggested to me that writing textbooks was no longer worthwhile when everyone could get their information from large language models (LLMs). As I always try to take a balanced view on LLMs—neither subscribing to the hype nor dismissing them out of hand—I feel the need to engage with this viewpoint rather than just writing books as we have for the last 30 years. As LLMs have improved, it’s clear that a query about a networking topic, which might have once led a student to the online version of one of our open source books, will increasingly (for better or worse) be answered by an LLM. Recent announcements by Google exemplify the latest efforts to minimize direct reader engagement with any source material outside an LLM.

At this year’s SIGCOMM conference there is a workshop on networking education, which is obviously of interest to us at Systems Approach given our focus on textbooks and other educational materials. We contributed to an earlier instance of this workshop, discussing our foray into open source textbook publishing, in 2020. The workshop this year has the somewhat provocative title “Networking Education for the AI Generation”. That has led to my thinking again about what value we can offer as authors in this brave new world.

A guiding principle of our work, going all the way back to our first edition of Computer Networks: A Systems Approach (1996), has been to provide the reader with a sense of perspective. One of our favorite reviewer quotes appeared on the jacket of our first edition, from Chris Edmondson-Yurkanan:

This book is incredibly rich in perspective! Each network protocol is first motivated by a unique description of the problem that needs to be solved, followed by a very clean abstraction of the important issues, and finished with a discussion of performance and implementation. I like this book as much as I like the Hennessy/Patterson architecture texts.

We don’t just love the comparison to legendary computer architecture pioneers and authors Hennessy and Patterson. This sense of perspective is what we were striving for in 1996, and we think it’s still a unique value proposition compared to what one gets from an LLM. Perspective means helping the reader see the bigger picture, and situating technologies relative to that big-picture view.

In those early days, one of our principles was: don’t be an encyclopedia. By that we meant: don’t assume that every networking technology or protocol needs to get its own paragraph or subsection. Help the reader grasp the important aspects of networking. Choosing what to omit is a part of that goal just as much as choosing what to include, because that tells the reader what’s central rather than peripheral.

Deciding what to leave out has always been a bit of a challenge with the revisions to our book: it’s painful to drop a section that you sweated over in the last edition. With each edition we’ve tried to be disciplined in dropping technologies that haven’t stood the test of time or that don’t serve an important purpose to illustrate some foundational principle. Our new edition is such a comprehensive refactoring relative to prior editions—essentially a clean slate approach—that it’s been much easier to drop material that doesn’t serve a solid educational purpose.

If “don’t be an encyclopedia” guided us in previous editions, “don’t replicate an LLM” is a reasonable goal for this one. Just as an encyclopedia has something on every topic, an LLM will give you an answer to any question (making something up if necessary). But how does the student of today know which questions to ask? That is where we hope our sense of perspective comes into play.

Building a Mental Model

One of the challenges to providing a sense of perspective was described in Larry’s "fitting it all in your head" post. With modern networks being so complex, how does a student get a mental picture of networking in their head? After a couple of iterations on our book outline with input from some friendly professors, we’ve settled on a 3-part structure for the book, with Part I aiming to solve this problem. It starts by defining the problem space that today’s networks need to address and then gives a top-to-bottom view of networking—necessarily leaving out a lot of details. Our thinking is that by covering key requirements such as scalability, cost effectiveness, resource sharing, and performance, and some core implementation artifacts like application protocols (exemplified by HTTP), packet switching (exemplified by Ethernet), we can get across a big picture of networking in Part I. In other words we are using a two-pass approach.

The rest of the book—the second pass—fills in the details. Part II focuses on what’s inside the network: routers, routing protocols, shared...