This is a thinking-out-loud post. There are no final answers here — only honest attempts. If you disagree, the disagreement is probably useful.
The Before and After Nobody Is Saying Out Loud
Five years ago, building a web application that could handle real users required understanding HTTP verbs, database normalization, state management, deployment pipelines, and at least one backend framework reasonably well. In most paths through the industry, that learning curve was measured in years, not weeks — though the honest range was always wide. Behind that curve sat a very real gate: only people who cleared it got to build meaningful things.
That gate is gone. Or rather, it's moved.
A person who can think clearly and describe a problem well can now produce working software in days that would have taken a junior engineer months to approximate. A non-technical founder can prototype their own product. A teacher can build a tool for her classroom. A logistics manager can automate a process that was eating three hours of his week. None of them need to understand what's happening underneath — and that is a genuinely extraordinary shift in human capability.
This should be celebrated. Seriously. The compression of years of learning into days of building is one of the most democratizing things to happen to technology in decades.
But celebration doesn't require pretending there's no other side to it.
The Engineers Nobody Will Ever Discover
The conversation about AI and developers is dominated almost entirely by people who are already visible. People who work at large tech companies, give conference talks, write newsletters. They speak with confidence about what the future of engineering looks like — because they've seen the most sophisticated version of it.
But there is a version of this industry that almost never appears in that conversation, and it is, in most measures, the larger one.
Somewhere right now, there is an engineer working for a mid-sized logistics company in a city nobody outside that region has ever heard of. She has built a database architecture that handles their operations with a reliability that would impress anyone who looked at it — but nobody has looked at it, because nobody knows to look. She doesn't go to conferences. She earns a salary that people at a Series B startup in San Francisco would consider embarrassingly low. She may be genuinely better at certain things than people earning four times her salary — not despite the constraints she works under, but partly because of them. Constraints produce ingenuity. Necessity produces elegance.
There are engineers like her everywhere. In small towns in Egypt, in regional cities in Brazil, in mid-sized companies across Eastern Europe and Southeast Asia. They solved real problems quietly, without recognition, without the social capital that turns technical skill into visibility. Many never got the chance they deserved — not because they weren't good enough, but because the systems that allocate opportunity are not particularly good at finding people who aren't already in the right networks.
What happens to these engineers in the AI era? One answer is optimistic: if output quality is what gets recognized, and AI helps produce higher-quality output regardless of background, maybe the hidden engineers finally surface. Maybe the playing field tilts toward people who think well rather than people who went to the right school.
But another answer is darker: AI further advantages whoever already has the most context, the most sophisticated prompting skills, the most access to the best tools — and that is still, disproportionately, the people who are already visible. Advantage compounds. It always has. Both answers might be true in different contexts, simultaneously. That is the honest answer.
The Survivorship Bias in the "Learn Fundamentals" Debate
Much of the loudest advice to "learn fundamentals" comes from people who cleared the old gate — who got into Google, Stripe, or similar companies, and built their careers there. That advice comes from genuine experience. But it is advice shaped by an extremely filtered dataset.
Most software work, in practice, has very little to do with distributed systems at scale, algorithmic complexity, or compiler design — though the engineers doing that work would have a better view of its real complexity than any outside observer. In most cases: CRUD applications. Admin dashboards. Inventory systems. Report generators. Internal tools at companies that have never heard of a Hacker News post. The engineers building those systems do it well, reliably, and without most of the "fundamentals" the online discourse treats as nonnegotiable.
Not everyone needs to work at Microsoft. Pretending otherwise — using the standards of the top 0.1% of companies to judge the entire industry — quietly diminishes the many engineers doing genuinely valuable work who are told, implicitly, that they are not real engineers.
The hypothesis worth sitting with is this: for most of the market, output has historically been rewarded more reliably than depth. Depth was always a premium that some employers paid for, in some roles, at certain scales. AI didn't invent that reality. It just made it more visible — and more uncomfortable for those who built their identity around being the person who understands.
The Identity Question Nobody Is Naming Directly
A significant number of engineers built their sense of professional self — their dignity — around being the person who understands. The person who can read the error nobody else can read. Who knows why the memory is leaking. Who sees the architectural problem before it becomes a crisis. Being the person who knows was the source of their worth, in a way that went well beyond the paycheck.
When AI generates code indistinguishable from what took years to learn to write — or better than it — the implicit question that arises is not "will I lose my job?" The deeper question, the one that produces the defensive reaction, is: If a machine can do this, what were all those years for?
That is not a technical question. It is an existential one. And it deserves an honest answer rather than a motivational speech.
The honest answer: those years built something real. They built a calibration — a felt sense of what problems actually are, what failure feels like, what correctness means under pressure. That calibration is not the same as the knowledge itself. You can forget specific syntax and keep the judgment that learning it trained. That judgment is what lets you recognize when an AI-generated solution is superficially plausible but structurally wrong — in a way that someone who never built it by hand often cannot.
But the market value of that calibration, as a standalone commodity, is declining. Not to zero. But declining. Pretending otherwise, in the name of encouragement, is doing a disservice to people who need to make real decisions about their time.
The Accountability Gap Nobody Is Building Culture Around
When someone uses AI to generate a payment module and ships it — and that module has a vulnerability that exposes user data — who is responsible? Legally, the company. But professionally, in the deeper sense: what does it mean to own something you didn't fully understand when you shipped it?
Aviation built an entire professional culture around exactly this problem — checklists, mandatory reporting, incident databases, regulatory oversight, black boxes. The culture was designed to match the stakes. Software never fully did that, partly because the consequences of failure were historically cheap and reversible. Push a fix on Monday. Nobody died.
AI is expanding what can be built, and how fast. The potential scale of failure is growing. And we are still operating with the cultural norms of an industry where failure was cheap. That mismatch is the real structural problem — not whether any individual developer goes deep or uses AI, but whether the profession is building the accountability infrastructure to match its growing capabilities. That work is happening too slowly, and almost nobody is framing it that way.
What Is the "New Life," Exactly?
There is a version of the AI future being sold right now: AI does the work, humans are freed. People can pursue creativity, leisure, meaning. Nobody needs to do the grinding, repetitive parts. The implicit promise is that the new life is better.
But there are two very different versions of that new life.
In one, people are freed from drudgery to do work that is more meaningful — more creative, more relational, more purposeful. This version requires a careful, intentional transition: new economic structures, new educational systems, new cultural values. It doesn't happen automatically.
In the other version — the one that tends to happen when economic incentives are left to themselves — the people whose labor is displaced don't find creative fulfillment. They find precarious service jobs, or enforced leisure that doesn't feel like freedom because it wasn't chosen. The "doing nothing" isn't retirement. It is a form of invisibility.
Work is not, in most people's experience, purely about income. It gives structure to time, identity, a sense of contribution, community. The relationships formed at work are often among the most significant in an adult life. Even if income security were fully solved — by policy, automation surplus, or anything else — the structure problem, the identity problem, the contribution problem, and the community problem would remain. A person who is materially secure but has no meaningful use for their capabilities is not flourishing. That much has been understood across philosophical traditions long before AI made it a practical question.
The question of what replaces work — not economically, but psychologically — is one of the least examined questions of the AI transition. It is also the question that receives the least serious attention in technology circles, perhaps because the people driving the technology are the ones whose work is not at risk.
A Thought Experiment: What If We Had Generative AI Ten Years Earlier?
Imagine it is 2015. Cloud infrastructure is established. The smartphone has matured. And in this version of history, an AI trained on every book, course, paper, and engineering manual ever written is freely available to every student on earth.
The optimistic answer is obvious: we accelerate everything. By 2025, in this version, problems that still exist in reality have been solved.
But there is a less comfortable possibility worth sitting with. Many accounts of major scientific breakthroughs share a structural feature worth noticing: the person who made the discovery was often working at the edge of existing knowledge, where the frameworks started to fail. Newton developed calculus in part because existing mathematical tools were insufficient for the problems he was trying to describe. The inconsistency between Newtonian mechanics and electromagnetism was visible to many physicists before Einstein; what distinguished his approach was sitting with that discomfort long enough to let it become a radical question rather than a problem to file away.
Discovery tends to happen in the gaps — in the places where existing frameworks break down and don't yet have an answer. If you have a tool that generates a confident, coherent answer on demand, the experience of sitting with a gap long enough to let it become a new question gets shortened or removed. Whether that makes the world more inventive or less is genuinely unclear. The 2015 AI might have made the world dramatically better at applying what it already knew. Whether it would have made the world better at discovering what it didn't know is a different question entirely — and one that deserves more thought than it's getting.
The Standardization Risk and the Schools Question
If everyone learns from the same system, everyone learns the same things in the same way with the same blind spots. That's not a criticism specific to AI — it applies to any standardized curriculum. But AI at scale makes it more acute.
The history of discovery is full of breakthroughs that came from unusual combinations — the biologist who also studied physics, the linguist who thought about information theory. These combinations happened because human learning is messy and non-standard. When a system optimizes for the canonical understanding of a field, it is very useful for bringing everyone to a baseline. It is potentially corrosive to the diversity of thought that produces the non-canonical idea — the one that doesn't fit, gets dismissed, and eventually turns out to be correct.
This connects to the schools question: did we learn wrong? Memorization is the most criticized aspect of traditional education — and the critique has real force. But memorization was never purely about storing facts. It was about internalizing: running something through your mind enough times that its structure becomes intuitive, becomes part of how you think rather than something you retrieve. That capacity — to internalize rather than just retrieve — is more important than ever. The question is whether we're finding new ways to build it, or letting it atrophy while calling that progress.
Now Is the Past — We Just Can't See It Yet
There is a disorienting exercise worth doing: assume that right now, 2026, is the past.
Not a distant past — the same distance forward as 2015 is behind us. Ten years. Imagine someone in 2036 reading a history of this moment. They see a generation that had access to genuinely transformative intelligence tools — and spent most of the early years arguing about whether those tools were legitimate. Debating whether "vibe coding" was real engineering. Having the same circular argument about fundamentals on different platforms, over and over.
From 2036, this will probably look the way the 2012 debates about whether smartphones were ruining attention spans look to us now. Recognizable in their anxiety. Largely missing the actual point.
But the person in 2036 might also see something else: the decisions made in these early years — about how to build professional culture around AI, how to design education for it, how to structure accountability — those decisions turned out to matter enormously. Not in the ways anyone predicted. The effects were slow and compounding, visible only in retrospect.
The people in 2015, at the beginning of the modern cloud and mobile era, were making decisions whose full consequences they couldn't see. We can see those consequences now. The same is true of us today — we are making decisions with incomplete visibility into what they mean at scale. Most of them are not intentional. They are the defaults chosen by companies optimizing for quarterly results, by educators working with outdated curricula, by engineers using whatever ships faster.
The question that cuts through most of the current noise is not "should I use AI or learn fundamentals?" It is: what are we deciding right now that we will not be able to see the full consequences of for another ten years?
Because those decisions are being made. By educators designing curricula. By companies choosing what to optimize for. By engineers deciding what to understand deeply and what to delegate. By parents deciding what relationship to let their children have with AI from the beginning. The children starting primary school right now will be finishing their education around 2036. Their relationship to work, to learning, to contribution, will be shaped by choices being made today — in ways we cannot fully model from here.
Where This Leaves Us
The vibe coder debate is ultimately not a debate about coding. It is a debate about what kind of world we are building — who gets to participate in building it, what it means to contribute, and what happens to human meaning when the most visible forms of human skill become easier for machines than for people.
Those questions don't have clean answers. But they have more honest answers than the ones currently circulating — which tend toward either defensive nostalgia or utopian optimism, both of which serve the comfort of the person saying them more than the clarity of the people listening.
The engineers who are undiscoverable — the ones quietly solving hard problems without recognition or adequate compensation — deserve to be part of this conversation. The workers whose labor is being displaced deserve a more honest account of what the "new life" actually looks like in practice, not in promise. And the profession deserves to ask itself whether it is building the accountability culture that matches the power it is accumulating.
These are harder questions than "should I learn Rust." But they are the questions that actually matter.
These are thoughts in progress. The questions feel more solid than the answers. If you're thinking about this differently, the conversation is worth having.