For the last ten years, software engineering ceased to be a calling and became a safe harbor for relatively smart people who wanted a six-figure salary. As long as you showed up and performed within one standard deviation of the mean, the market guaranteed you a cushy existence. Corporate tech became a mechanism for distributing capital to the moderately competent.
That era is over. We are shifting from a labor market defined by normal distributions to one defined by extreme Power Laws.
The Physics of Talent
Most jobs are bell curve jobs because output is naturally capped. A hotel clerk can't check in 500 guests simultaneously; the best clerk is only modestly better than the average. In these types of roles, the best strategy is to hire "good enough" people and to scale by adding more.
But high-leverage domains aren't capped the same way. Acting, venture capital, pro sports—these are all power-law arenas. A handful of outliers are responsible for almost all value, and the average participant's contribution becomes irrelevant.
For a long time, software looked like a blend of both models. You had high-variance coders, but the gap between the mean and the +2σ/+3σ engineers wasn't wide enough to completely overshadow the rest. Even if 2% of engineers shipped 25% of the total value, the remaining 75% of the work still needed to get done.
Agentic Engineering changes the physics. The "extra hands" just stop mattering.
The Great Bifurcation
AI will give software engineers the biggest year-over-year productivity increase any industry has ever seen, but it won't be harnessed evenly. The biggest mistake anyone could make in 2026 is to imagine a uniformly applied multiplier.
Let me give you an imperfect analogy: imagine aliens dropping into 15th-century Europe, bearing gifts for a select group of 100 farmers. 90 get a small tractor, 9 get a combine. The last one gets a factory that builds driverless combines.
In the above scenario, no one would dispute that all 100 farmers became dramatically more productive. Still, 99 families are packing up for the city because their individual marginal contribution rounded down to zero anyway.
That's what's happening to code: the leverage gap will get so wide that the middle turns into a coordination tax.
Should I Stay or Should I Go?
This is a thought experiment for currently employed engineers thinking about their future.
Imagine your boss walks into the office on Monday and tells you and the 39 other engineers on your team: "Two weeks from now, 30 of you are gone, and the remaining 10 will have their salaries tripled."
Do you feel excited—or do you panic?
If the honest answer is panic, you only have a few months left to build your metaphorical driverless combine factory. The "lifestyle" tech job is dead. If you don't really love this stuff, if you are still coasting, you are foolishly shorting the AI market.
What About Employers?
Assuming you accept my premise, realize that the hiring playbook just changed.
You have to start building engineering teams like a baseball scout. In a power-law world, the job of hiring isn't to avoid false positives. It's to find the outliers.
The contributions of +2σ/+3σ people will dwarf everybody else's.
That means you should short-circuit the interview process fast. The moment it's obvious a candidate is a good, "95 mph forever" profile, move on. Scouts don't bet on the player who throws 95 with perfect control. They bet on the kid who might touch 106 one day. Even if he's raw. Even if he's weird. Even if he doesn't interview well. Because two years from now, the 95 mph guys are on the bench anyway—and the only thing that matters is whether you found the arm that breaks the game.
In this new world, a milquetoast positive reference check is probably a strong signal that the upside is capped. Do not hire. Conversely, a bad review that uses the word "difficult" might just be code for unreasonably high agency.
The Only Thing that Matters
For engineers, chasing 2+ sigmas or leaving the profession are the only two rational choices.
Eighteen months from now, for every developer who's better than you, there must be dozens who are obviously worse or you are unemployed. Remember, the very best baseball player you went to high school with is probably an accountant.
Raw talent alone won't get you there. The only reliable path is aggressive iteration: run experiments, try weird workflows, burn tokens on dead ends, and keep the handful of things that actually move your slope. If your employer doesn't support that—budget, permission, cultural air cover—the rational move is unfortunately to find one that does. Whether you take the rational move or not is entirely your call.
If you love writing all the code yourself, I'm sorry, but it's time to give it up.
The Only Thing That Matters: Employer Edition
If you're the employer and you're lucky enough to have a few of these weirdos on your team, you can't afford to lose them because by 2028, they'll be writing all the code. You might have to give them more free rein that you historically would have, sponsor their side projects, or just pay them more than anyone else while they figure all this out. But you have to keep them.
Obviously, I might be completely wrong. However, the cost of over-indexing on the bifurcation of talent is a few thousand dollars in wasted tokens. The cost of under-indexing is a department full of useless people.
If this turns into a baseball economy—like I believe it will—you need to have the pitcher who can throw 105 mph fastballs on your roster. Replacing the superstar with ten guys who throw 95 wins zero games.