The first assignment

In 1996 I started as a field network engineer intern at Cabletron Systems in Brazil. The internship became a four-year job and then a career, but on the first day the assignment was smaller than any of that: read this book, all of it.

The book was The Cabletron Systems Guide to Local Area Networking, published in 1992. It is a paperback with a cartoon on the cover of a man in a shirt and tie, sleeves rolled up, carrying a binder under one arm and reading another. I still have it. The cover is scuffed white and the spine has gone soft, which is what happens to a book somebody actually read.

Its introduction is disarmingly honest about why it exists:

Today we are all facing the same problem, the lack of skilled staff, specifically in the Local Area Networking area. Yesterday you just managed to get to grips with the difference between a bridge and a router and up pops a brouter. If I am already part of the industry and I am having trouble then what chance have your new staff?

And in the margin beside that paragraph, next to a drawing of a man collapsed in an armchair surrounded by fallen books:

There are not enough hours in the day to do the job and keep up with the technological changes.

That was the industry's self-description in 1992. I would not change a word of it to publish today.

The contents page

Nineteen sections. Some of them are exactly the archaeology you would : Section 4, Introduction to FDDI. Section 5, CuDDI - The Copper FDDI Solution, which notes that the standard "is close to ratification" and warns about a non-standard "Green Book" group. Section 14, Networking in a multiprotocol solution. Section 15, which states that SNA networks account for fifty per cent of the networks installed. That number is the most 1992 sentence in the book.

Others have not dated at all. Section 2, The Cable is the . Section 3, and Review - one of those two is still under my students' desks. Section 9, Wireless LANs, which says networks without cabling are now possible but bandwidth availability is slowing uptake, and that there are nonetheless some applications for which wireless is the correct solution. Section 10, Network Management - A chance to get it right. Section 16, A Network Strategy based on Fundamentals not Fashion, which is a better title than most of what I have read this year.

And then Section 12.

"Network Management based on Artificial Intelligence"

Here is the entry, in full, from the contents page of a book published in 1992:

Following on from Network Management there is a major move to introduce Artificial Intelligence into managing a network. This section discusses the option of Rules Based AI against Inductive Modelling Based AI.

A major move. In 1992. With a taxonomy already: rules-based on one side, inductive modelling - learning the network's behaviour from its own data rather than being told the rules in advance - on the other. That is not a footnote or a speculative afterword. It is section twelve of nineteen, sitting between and RMON on one side and hub technology on the other, treated as an ordinary part of the subject a new hire was expected to learn.

I read that in 1996 and, if I am honest, I remember nothing about it. I was twenty and the sections I needed were the ones about cable.

What I do with that now

I spend my working week teaching security and networking platforms whose current marketing rests entirely on the second half of that taxonomy. Behavioural analytics. Learned baselines. Anomaly detection that does not need a rule written in advance. Every I am authorized to teach ships some version of it, and every one of them presents it as new.

It is not new. It is section twelve.

I want to be careful here, because there is a cheap version of this observation and I do not want to make it. The cheap version says nothing ever changes and everyone is selling you the same thing in new packaging. That is false, and anyone who has watched a modern behavioural engine catch something a rule set would have missed knows it is false. The models are enormously better. The data available to them is enormously better. The compute is not comparable. Something real happened between 1992 and now.

The honest version is narrower and, I think, more useful: the argument has not moved. The 1992 book framed it as rules against inductive modelling, and that is still the framing. Every practitioner I teach is still choosing, in practice, between a control they can read and predict, and a control that performs better but cannot tell them why. The failure canon on this site is full of systems that did exactly what they were configured to do, and the reason a rule set survives in production alongside a smarter engine is that when the smarter engine is wrong at three in the morning, somebody has to be able to say what will happen if they turn it off.

That is not a 1992 problem or a 2026 problem. It is the problem, and the book named it before I could drive.

What the book is actually evidence of

Two things, and they pull in opposite directions.

The obsolescence is real and it is total. CuDDI did not happen. FDDI is a museum piece. The "Green Book" group is not even a memory. SNA fell from half of installed networks to a curiosity. Almost every specific technical fact in that book is now wrong, and I have written an article on this site about how badly this industry predicts which things will die and which will not - it declares finished what outlives the declaration by decades, and treats as permanent what vanishes inside five years. This book is a full worked example of both errors in one volume.

And the shape is entirely intact. Not enough hours in the day. Not enough skilled staff. A standard close to ratification with a non-standard faction. Fundamentals against fashion. Rules against learning. Those are the same conversations I have with students in 2026, and they were already old when this book was printed.

Which is why I keep it, and why the assignment was a good one. A book of facts that are all wrong turns out to be an excellent way to learn which parts of the job are facts and which parts are the job.

The part I did not expect

I have been doing this for thirty years, and until recently I would have told you the through-line was the technology. Reading that contents page again for this piece, I do not think it is. What the 1992 book was worried about, in its introduction and its margins and its cartoon of a man buried in his own reading, was people: that there were not enough of them who knew this, that the ones who did could not keep up, and that somebody had to teach the new ones.

Cabletron's answer was to write a book and hand it to an intern.

Thirty years later that is still, more or less, my job.