GF.

The Death of the Nerd

For decades, being brilliant at the technical part bought you an exemption from everything else. AI just revoked it.


Every tech company used to have one. The engineer who answered in monosyllables, dodged every meeting he could, and let the product manager translate him for the rest of the world. Nobody minded. He was brilliant, and brilliance was the deal. The industry ran for decades on an unwritten contract: be exceptional at the technical part and we will forgive everything else. The PM translated, the manager sold, a colleague gave the demo. His code spoke for him, and it spoke loudly enough that he never had to.

I grew up professionally inside that world. I was never its most extreme case, but it is my tribe, and I am watching the contract that protected it get revoked in real time.

What changed is the execution layer of software. AI compressed it. I see it in my own work: things that used to need a team, a sprint and three alignment meetings now need an afternoon and a well-written brief. And the compression does not stop at engineers. People who never wrote code are shipping real products to production, and I mean solid ones, not toys. Google says three quarters of its new code is now AI generated, and one of the biggest of the new app builders reports that 80 percent of its users do not consider themselves technical. But I did not need those numbers to believe it. I have seen it happen up close.

Yes, some of it blows up. Leaked keys, deleted production databases, founders discovering security the hard way. The floor is not zero, and pretending otherwise makes the argument worse, not better. But none of these failures restores scarcity, and scarcity is what the contract ran on. It rested on how few people could produce software, not on how hard software is. That number moved by an order of magnitude, and a contract backed by scarcity does not survive abundance.

The obvious objection is that someone still has to write the compiler, the database, the payment rails, the models themselves. True, and worth taking seriously, because that is exactly where the pure technical profile went. The refuge is real, and the frontier labs pay salaries that look like typos. But count the seats. Entry level hiring at large tech companies has fallen by more than half since 2019, and the middle of the market is where our uncommunicable nerd lived. He was rarely the person writing the database engine. He was the person three floors up, writing application code behind a PM, protected by the fact that application code was hard. It is not hard anymore. The refuge exists, but it does not fit everybody, and it was never going to.

So the bottleneck moved. If technical execution is compressed, what remains expensive is everything around it: figuring out what deserves to be built, talking to the people who will use it, convincing someone to pay, aligning the humans who have to say yes. Every one of those is communication, the kind that moves an accurate picture of reality from one head to another, not the slideware kind. Read a handful of engineering job postings today and it sounds like a personality transplant prescription for my tribe: stakeholder management, public speaking, time with customers.

What made me write an essay instead of a tweet is the asymmetry. Two profiles are racing toward the same job, the hic (High Impact Contributor), the person who can run the whole process end to end, and they are not starting from the same distance. From one side comes the communicator learning to build. They have to learn to specify, to decompose a problem, to ask precisely. That takes months, because the machine meets them more than halfway, and their native skill, articulating intent in plain language, turns out to be the interface. Karpathy called it years ago: the hottest new programming language is English. From the other side comes the uncommunicable nerd learning to communicate. He has to rebuild a capacity he spent a lifetime routing around, and he has to do it in public, with feedback, against discomfort. That takes years. Both conversions are possible, but only one of them runs uphill. The contract is expiring, and it is leaving the tribe that depended on it with the harder half of the retraining problem.

If you recognize yourself in any of this, the prescription is short, because it is not complicated, only uncomfortable. Communication is a skill, not a gift. It responds to training like any other skill, and the technical person who trains it keeps an advantage the converted communicator cannot easily buy: the judgment to know what good work looks like and when the machine is confidently wrong. The introvert even gets a head start, because the natural gym for this is writing, and writing is the introvert’s home field. The muscle that writes a clear spec for a model is the same muscle that writes a clear argument for a human.

References

  1. Fast Company, “Google CEO says 75% of the company’s code is AI-generated” (April 2026)
  2. The Next Web, “Lovable hits $500M ARR as the ‘build economy’ takes off” (June 2026)
  3. SignalFire, State of Tech Talent Report 2026, entry level hiring data (2026)