We've been told since late 2022 that "within 6 months, we won't need software engineers anymore". I think that's half-right.
I also think it's jus...
For further actions, you may consider blocking this person and/or reporting abuse
I think the core idea here is really about adaptability under changing systems rather than AI specifically.
The T/comb framing makes sense because most real work now sits across boundaries, not inside a single silo.
Though I’d still argue depth doesn’t disappear — it just becomes the anchor that lets generalists stay correct while moving across domains.
Yeah, in retrospect I probably would've added that this isn't unique to the AI hype bubble, but just the most recent iteration of "how do we deal with change."
The AI-pocalypse is just so... in our faces right now... everybody selling us that this is a watershed moment in history where EVERYTHING IS UNPRECEDENTED... but eh. It's the next verse of the same song.
I mostly agree with the “same song, different verse” framing—but I think there’s a subtle twist worth pulling on.
If we assume AI really does compress a lot of “I-shaped” execution work (syntax fluency, boilerplate, known patterns), then the surviving “depth” might shift upward rather than sideways. In other words, depth in problem framing and constraint understanding becomes more valuable than depth in a specific tool or language.
That creates an interesting tension in your T/comb argument: generalists win on adaptability, but specialists don’t just vanish—they get redefined. The “I-shaped” engineer who survives is probably not the React expert or the Kubernetes expert, but the person who deeply understands distributed systems tradeoffs, security models, or user behavior under constraints—things AI can describe but not reliably own in context.
So maybe the real split isn’t I vs T/comb, but:
replaceable depth (implementation-level expertise)
irreplaceable depth (judgment-heavy, ambiguity-heavy domains)
and connective tissue (the T/comb layer)
Which leads me to a question back to you: do you think AI pushes more people into becoming “connectors” (T/comb), or does it actually raise the ceiling on what counts as valuable “deep” work and preserve strong I-shaped roles—but only in narrower, more abstract domains?
Personally I'm of the opinion that a lot of the AI hype is going to crash and burn when people realize it can't literally solve all their problems (or it's made economically infeasible as we're already seeing it to shift). If it settles into its niche ok, I see the connectors being more valuable because one use case where AI does a great job is in training in depth... there might be a few I-shaped roles left out there but they'll be the absolute deepest I's ever who survive... the ones whose knowledge takes a lot longer to surpass in the retraining loops.
T's and Combs have an easier time of things because they think more laterally.
So maybe in summary, I see MORE people being pushed to connector roles but also the truly deep I's will go deeper... maybe with just fewer of them around.
I think you’re right that AI squeezes out the mid-depth “I-shaped” work first, while pushing more people into connector roles.
The interesting part is the remaining true deep experts don’t become obsolete-they become rarer and more important, because their knowledge is harder to fake or compress.
So the system ends up more polarized: lots of connectors, fewer but deeper specialists.
Do you think connectors eventually become the default career path, or just a transition layer people pass through?
I don't know about default - career paths are weird things and so much depends on what people like to do. Connectors and Deep Experts seem like parallel paths that (at some point in a career) may transition into the other.
Great point of view. This debate has been going on and will go on ( probably). You talked about different skill shaped people. Any advice on someone is currently on the learning stage. The AI hype has been going around so much that it sometimes feel more confusing when everybody are saying to focus on strengthening your fundamentals. If you were asked What would say about what exactly are these fundamentals?
The fundamentals are things like Collaboration via Source Control, CI/CD delivery automation, making quality tests a first-class citizen of your process, and thinking about your code from a Security perspective.
These concepts transcend the topic of a certain language or technique and address how the work gets done... the same fundamentals can be applied regardless of what you're buildng.
If you're in the learning stage, those are the things I would learn FIRST - from there, you can pick up any language or technology and you have the tools to manage what you build easily. Note that you can totally get AI to help with these things, but you need to know how to do them manually first.
full-stack has been winning in small shops forever - the AI wave just accelerates the squeeze on big-company silos. what I’m curious about is whether ‘full-stack’ means the same thing in 5 years as it does now.
The generalist-wins thesis holds for a reason worth naming: AI collapses the cost of any single specialty, so the scarce skill becomes connecting them, knowing which specialty a problem needs and whether the generated piece is right. A specialist competes with the model on the exact axis the model is strongest. A generalist competes on judgment across boundaries, which is the part the model is weakest at, because it has no stake in whether the pieces fit. The later-instead-of-sooner caveat is fair though: the transition punishes generalists who are shallow everywhere and rewards the ones deep enough in each area to catch the model when it's confidently wrong. Breadth without enough depth to verify is just faster trust in output you can't check.
AI may compress the advantage of narrow, isolated expertise, but it doesn’t remove the need for engineers who understand context, systems thinking, and real-world trade-offs. If anything, it raises the value of people who can connect dots across layers, not just produce code in one.
The real shift feels less like “AI replacing engineers” and more like “AI amplifying those who can think beyond a single layer of the stack.”
❤️
Great article. Thanks for sharing your experience.
I'm currently exploring Claude Code, Codex and AI agent workflows. This was helpful.