By Victor Zhong, May 13, 2026
In the analog-computing gallery of the Mountain View Computer History Museum, there is a photograph of Bob Widlar captioned with his line, "Digital circuits? I'd rather fight than switch!" Widlar was among the most brilliant and eccentric analog integrated-circuit designers of his generation, the man behind some of the first widely successful monolithic operational amplifiers. The line is in character: he spent his career, by most accounts, refusing to switch.
He was, of course, wrong about digital. The next several decades belonged to it, and the analog computers in the gallery around his photograph are now museum pieces - though the analog IC industry he helped found is larger than ever. The picture stays with me anyway. Widlar held out because he believed his paradigm was the field. The people who actually carried computer science forward across that transition were the ones who could see both rooms at once: trained broadly enough to recognize that the questions had not changed even if the substrate had.
I am a professor of computer science, and I came into the field sideways. My parents were electrical engineers. When I was seven, they gave me a box of SnapCircuits, a children's kit of resistors, inductors, and capacitors that you assembled according to a manual. I built doorbells and radios. Years later, in my first circuits lab at the University of Toronto, my main thought was: this is just like SnapCircuits. I had wanted to be anything but an electrical engineer like my parents; I had watched An Inconvenient Truth and wanted to do something about climate change. I ended up in power systems, then in FPGA design, then at an internship at Altera in San Jose, where I first heard about artificial intelligence, and then at Stanford, where I did a master's in the NLP lab. It was a meandering path, and the field that received me at the end bore little resemblance to the one I had entered.
Today, that path is largely closed. An undergraduate who walks into a computer science department in 2026 does not walk into the room I walked into. They walk into a room oriented almost entirely around artificial intelligence - its course catalogs, its funded labs, its job listings, its student culture. The path I took is not so much discouraged as invisible. And the field is poorer for that, in ways that are easy to see and harder to name.
The narrowing operates as a feedback loop. Students orient toward AI; departments follow. AI itself narrows, until the database conference and the systems conference begin to look like AI conferences in disguise. Recent SIGMOD programs have been heavy with sessions on LLMs for query optimization; the systems conferences tell a similar story. The image the next cohort forms, when they decide what to study, is of a field with only one good idea in it. They choose accordingly, and the loop tightens.
None of this is to devalue LLMs or transformers, or the people doing serious work on them. I work on agentic systems; the ideas are real, and the moment is real. What I have grown wary of is the implicit claim, embedded in everything from course design to funding, that this is the only moment - that the history of the field before the transformer was a long prologue to it.
The alternative is visible a few steps from Widlar's photograph, in the rest of the museum. Computer science was not a discipline until the 1950s. The first computers were physical objects you maintained the way you maintained a furnace; the discipline that came to be called computer science grew out of the slow shift from circuits to interfaces, from interfaces to algorithms, from algorithms, in many corners, to learning. Each move was made by people trained broadly enough to recognize old questions in new form. Attention is, after all, a data structure. Retrieval is a database problem. Reinforcement learning is a control problem. The discoveries that made AI possible were not made by AI specialists.
What can a professor do about this? Less than one would like. The incentives push the other way. The job market rewards a legible AI specialization more than it rewards breadth. Conferences reward submissions that resemble the median paper in the room. Funding gravitates to the keywords on the latest call. A student who spends a semester on programming languages instead of cranking out a third paper for a NeurIPS workshop is taking on real career risk. None of that goes away because anyone writes an essay.
But the small choices that compound into a career are made one at a time. A student can take the detour into a systems course. An advisor can decline to measure a student's year by paper count. A funding committee can read a proposal without the word "agents" in it and not flinch. A program chair can build a session that does not have "LLM" in the title. None of these is heroic. Each of them, made enough times, is what holding the field open looks like.
The homogenization of computer science is not, in the end, a problem about AI. It is a problem about how a field treats its own ideas - whether the current frontier is read as the whole map, or as one well-lit corner of a much larger territory. I would like my students to leave my lab knowing that the corner is bright and the territory vast. I would like them to keep the habit of curiosity, which is not a phase of training to be optimized away. It is the thing that made the field worth joining. It is also the thing that will let them recognize the next transformation when it arrives.
I think of Widlar not as a cautionary figure but as a kindred one. He cared about a way of thinking, and he fought for it. The discipline we now call computer science was built, in no small part, by people who took the question of what computing should be seriously enough to argue about it.