I was on the phone with a friend at Meta for an hour last Tuesday. He was breathing fast, the way people do when they're trying not to panic. The layoff rumors were everywhere. Internal Slack channels had gone quiet in that specific, ominous way. And he kept asking me the same question: "Is it true? Are we actually obsolete?"
I told him no. But I also told him the truth is uglier than the lie.
The Panic Factory
Let's start with the hypocrisy, because it's breathtaking.
Company A spent all of last year declaring software engineers are obsolete. Their recruitment portal is still wide open. Company M announced mid-level engineers would be completely replaced by end of year. Didn't happen. Meanwhile, an army of AI influencers is manufacturing panic on schedule—every six weeks, a new "coding is dead" thread to sell the course.
So is it all fake news? Not exactly. Because the layoffs are real. Brutal. Ruthless. Happening right now.
The question is: who actually ordered them?
The VCs Demanded Blood
I remember 2020. The pandemic hit, and the tech company I was with instantly cut 20% of engineering. No performance reviews. No project completion analysis. Just severance packages and deactivated accounts.
I took a senior executive out for coffee and asked why. His answer was blunt: "Because the VCs wanted it."
That's it. That's the whole reason.
Six months later, those same companies were in a frantic bidding war to hire engineers back—at higher salaries, with signing bonuses. The talent they discarded like broken furniture was now premium inventory.
We're watching the exact same cycle. VCs and institutional investors are intoxicated by the media narrative. They believe AI can replace headcount. More dangerously, they believe that if a portfolio company doesn't execute massive layoffs, the CEO is failing to implement AI efficiently. To a VC, failing to cut headcount means you risk losing your next round.
Are they right?
Here's what I've learned since 2020: for a fund managing hundreds of billions, your company is a single playing card. Whether the layoff thesis is technically accurate doesn't matter. If the entire capital ecosystem decides a wrong thing is right, it becomes reality. Read Foucault if you want the philosophy. Read the news if you want the proof.
The executives running these companies have no choice. They must appease the capital. But if the capital is wrong—and history says it is—then we need to ask the foundational question: Can AI actually replace software engineering?
What an LLM Actually Is
Strip away the agents. Strip away the RAG pipelines, the memory systems, the fancy tooling. What's left?
At its foundation, an LLM is a machine performing statistical reconstruction.
Anyone who's raised a child understands the difference. You can tell an LLM: "Be careful with the glass, it will break." It ingests billions of data points proving that dropping glass equals shattering, and outputs the correct statistical response. But a human child has to drop the glass. Hear the shatter. Maybe step on a shard and bleed. That's how they comprehend the physics and the consequence.
"But statistical output is good enough for coding," some will say.
No. It is nowhere close to enough.
The Constraint of Consequence
Because an LLM only understands statistical proximity, it is entirely incapable of breaking constraints based on consequence.
An LLM doesn't get woken up by PagerDuty at 3 AM because production is down. It cannot be sued by a client. It doesn't have to look an angry CEO in the eye and explain why the migration failed. It doesn't go to prison if an audit uncovers fraud. It doesn't feel the cold terror of seeing a $500,000 AWS bill caused by a single poorly optimized loop.
Consequences are human. And necessity—born from consequence—is the only force that drives true architectural leaps.
Think about the pre-AI era. Have you ever spent three months scouring StackOverflow and GitHub, found absolutely nothing, and been forced to invent an entirely new architectural concept? An LLM cannot do this. It has no reason to force a paradigm shift. It has no pain.
Let me give you the clearest example I know: asynchronous programming.
The Invention of Async
Imagine a universe where async/await doesn't exist. Everyone writes synchronous code. If you ask an LLM to "make this faster," it does what it does best: pattern interpolation within its training data. It suggests faster CPUs. More RAM. Database indexing. Tighter loops. It optimizes endlessly inside the constraints of the Synchronous Universe.
But a human engineer? She's feeling the agonizing pain of a blocking request locking up the entire server. Users are timing out. The CEO is in her Slack. She's bleeding money by the minute. And suddenly she stops and asks: "Wait... why does this have to happen right now? Why are we blocking the thread at all?"
She rewrites the constraint because she feels the pain. The LLM doesn't feel the pain. The LLM doesn't even feel the cost of its own token consumption.
Async wasn't invented by optimization. It was invented by desperation. By a human who needed to escape a trap that statistical improvement couldn't solve.
That's the gap. That's the entire game. And it's not closing.
The Correction
Once this hysteria settles—once VCs realize that statistical optimization cannot replace architectural consequence—the demand for high-level software engineering will violently return.
The serious question isn't whether engineers are obsolete. It's whether you can survive financially and professionally until the market corrects itself. Because the correction is coming. It always comes. The people who get fired now will be hired back later at a premium, just like 2020, just like every cycle before it.
But the ones who don't survive the gap? They weren't replaced by AI. They were replaced by panic.
The Lens That Makes No Sense
I'll close with something completely unrelated.
The header image for this post was shot using a simulation of the Leica Thambar-M 90mm f/2.2 lens. It's a $6,500 piece of glass that deliberately takes unclear photos. Soft focus. Dream-like aberrations. Flawed on purpose.
Statistically, it makes absolutely no sense. Every training dataset says sharpness equals quality. An AI would never invent this lens because the data says it's wrong.
But humans pay for it. Because it makes them feel something.
Never underestimate the gap between statistical perfection and human reality. The VCs are optimizing for the former. The future belongs to the latter.
— James, Mercury Technology Solutions, Hong Kong, May 2026
