Something Big Is Happening — But It's Not What You Think

Matt Shumer’s essay went viral because it said out loud what a lot of people feel: something massive is happening with AI and most of us aren’t ready. I agree with the feeling. I disagree with the framing. What we’re living through isn’t an extinction event for knowledge work. It’s an abstraction layer shift — and if you’ve been in tech long enough, you’ve seen this movie before.

I. The Pattern Nobody Talks About

Assembly replaced hand-toggled machine code. C replaced Assembly. SQL replaced hand-written file I/O. Cloud replaced hand-racked servers. Every single one of these transitions killed the old workflow. Every single one of them made the people who adapted more productive, not less employed.

AI is the next layer. It doesn’t replace thinking — it replaces the mechanical translation between intent and implementation. You used to write loops; now you describe what you want. You used to format reports; now you describe the insight you need. The cognitive work didn’t vanish. It moved up.

The viral essay frames AI as a force that will “do your job.” But abstraction layers have never worked that way. They eliminate the previous way of doing the job, not the job itself. The demand for practitioners didn’t shrink after C replaced Assembly — it exploded, because the productivity gain expanded what was possible to build.

II. The 90/10 Problem

Here’s what the essay gets wrong by omission: AI handles about 90% of a task brilliantly and then falls on its face for the last 10%. That last 10% is where actual expertise lives — production context, edge cases, institutional knowledge, the judgment call about which tradeoff to make when the data is ambiguous.

I work with AI daily, building agentic systems at scale. The models are astonishing at generating first drafts, boilerplate, and plausible-looking output. They are terrible at knowing when they’re wrong. They can’t tell you that the database schema changed last Thursday, that the regulatory requirement in market X differs from market Y, or that the customer’s real problem isn’t the one they described in the ticket.

The 90% is genuinely transformative. But treating AI as if it will handle 100% is how you ship bugs to production, publish hallucinated data, and erode the trust that took years to build.

III. “But This Time It’s Different”

Every abstraction layer shift triggered the same prediction: this one will be the one that makes humans obsolete. When compilers arrived, people said programmers were done. When databases shipped with query optimizers, people said DBAs were done. When the cloud automated infrastructure, people said sysadmins were done.

What actually happened was that the bar moved. You couldn’t compete as a programmer if you only knew Assembly after C existed. You couldn’t compete as a DBA if you only knew flat files after SQL existed. The practitioners who refused to learn the new layer did lose their jobs — but to the new version of their role, not to the tool itself.

The viral essay asks: “What happens when AI can do everything a junior developer does?” The historical answer is: juniors learn to work with the tool and do what used to take a senior. That’s not a threat — it’s a career accelerator.

IV. The Expertise Paradox

Here’s the part that’s counterintuitive: AI amplifies the gap between experts and novices rather than closing it. A senior engineer using AI writes better code faster because they know what “good” looks like. They catch the hallucinations. They know which suggestions to accept and which to reject. They use AI as a power tool, not a replacement for judgment.

A novice using AI produces plausible-looking output that may or may not be correct, and they can’t tell the difference. This is exactly what happened with every previous abstraction layer. SQL made expert database designers more productive; it also made it easier for novices to write queries that destroyed performance. The cloud made expert architects more powerful; it also made it easy for novices to accidentally spin up six-figure monthly bills.

The tool is an amplifier. It amplifies competence and incompetence equally.

V. What the Essay Gets Right

I don’t want to dismiss the essay entirely. It nails one thing: the speed of change is real and the psychological weight of it is heavy. People are anxious, and the anxiety isn’t irrational. Entire workflows are being rebuilt in months, not decades.

The essay is also right that ignoring AI is not an option. If you’re a knowledge worker and you aren’t actively experimenting with these tools, you’re making the same mistake as the Assembly programmer who refused to learn C. The tools won’t replace you — but a person who uses the tools effectively will outperform you.

Where the essay goes off the rails is the leap from “this is fast and disruptive” to “this is an existential threat.” Speed of disruption and nature of disruption are different things. The automobile disrupted horse-drawn transport very quickly. It didn’t eliminate the need for people who understood logistics.

VI. The Agency Illusion

The viral framing implies that AI has agency — that it’s coming for your job, like a competitor. This anthropomorphization is seductive but wrong. AI is a tool. A remarkably powerful tool, but a tool nonetheless. It doesn’t want anything. It doesn’t strategize. It generates statistically likely continuations of input sequences.

When we project agency onto AI, we make two mistakes simultaneously. We overestimate what it can do autonomously (it can’t handle the 10%), and we underestimate what we bring to the table (the judgment, context, and taste that makes the 90% actually useful).

The essay’s dramatic framing — “something big is happening” — is better served by a more precise observation:

. That’s not a doomsday prediction. It’s a career development roadmap.

VII. What to Actually Do

If you’re a practitioner reading this and feeling the anxiety, here’s what I’d suggest based on building AI systems in production every day:

Learn to verify, not just generate. The most valuable skill in an AI-augmented world isn’t prompting — it’s knowing whether the output is correct. That requires domain expertise, not prompt tricks. Stay technical: understand the systems deeply enough to know when AI is right and when it’s confidently wrong.

Invest in the 10%. Deep expertise in your domain — the edge cases, the production gotchas, the institutional context — is what AI can’t replicate. Double down on the knowledge that’s hardest to commoditize.

Use AI as a collaborator, not an oracle. The best results come from treating AI output as a first draft that needs expert review, not as a finished product. This isn’t a limitation — it’s the same workflow every previous abstraction layer enabled.

VIII. The Bottom Line

Something big is happening. But it’s not what the viral essay suggests. We’re not witnessing the extinction of knowledge work. We’re witnessing the latest abstraction layer shift in a pattern that’s as old as computing itself. The Assembly programmers who learned C thrived. The sysadmins who learned cloud thrived. The practitioners who learn to work at the AI layer will thrive.

The ones who won’t thrive are the ones who either panic and freeze, or who dismiss the shift entirely. Both responses are wrong. The correct response is the same one that’s worked at every previous transition:

.

Every abstraction layer shift looked like the end of an era. Every one turned out to be the beginning of a bigger one. This time is no different — not because the technology is less powerful, but because the pattern is exactly the same.