Skip to content

AI 2025-2026: My Falsifiable Forecasts for the Next Two Years

No one actually knows where AI is taking us, least of all the people who claim they do. But if you want to orient yourself in the fog of hype, panic, and Science magazine press releases, you have to try. Here are my falsifiable predictions for the next 12 and 24 months—a roadmap for both victory laps and post-mortems.

If I’m wrong, you’ll know it for sure, because I’ve tied my predictions to public indicators. The premise: We’re veering toward one of two paths, and there’s a vanishingly small middle ground. Either the “doomer” scenario—runaway productivity wipes out a swath of the economic middle—or the “progress” scenario, where the tools that threaten to hollow out work instead multiply it. I expect progress, but the doomer path is plausible. Let’s explore both.

The Doomer Scenario: The White Collar Recession Arrives

Let’s get the dystopia out of the way.

The hypothesis: Over the next two years, general-purpose AI is “good enough” at nearly all routine white-collar work. Most work doesn’t go away entirely, but it gets de-skilled, deskilled, and underbid. The creative flourish you used to put in your Excel sheet? It’s now worth as much as a smile at the GM plant in 1992.

How Do We Get There?

  • Coding agents aren’t fully autonomous, but that’s good enough
    Companies don’t need AI that can ship the new Twitter itself. They just need AI that can turn a normal developer into a 5x developer, or (crucially) turn a non-developer into a 1x or 2x developer.
    This creates the coding equivalent of call center jobs: armies of “AI supervisors” gluing apps together, reviewing AI-generated pull requests, and getting paid a wage that slowly drops every quarter.

  • Productivity monitoring and the new Taylorism
    When every click, slack message, and line of code can be monitored, the performance management playbook writes itself. In high-supply, low-wage scenarios, white-collar work becomes the new fulfillment center.
    Managers have already figured out that you can use AI to penalize (even fire) workers for committing the professional equivalent of using the bathroom too often—dipping in productivity, flagging coverage on the ticket queue, not writing enough lines of code. The fantasy of “bring-your-whole-self-to-work” dies with a whimper and a Figma plugin.

Falsifiable Markers of the Doomer Slide

Here’s how you’ll know we’re living in this future:

  1. Tech job postings collapse—per Indeed/FRED, below 50, compared with a historical baseline of 100
    If you think it’s all just a cold spell, look for what happens to this number next spring.

  2. Mass layoffs in the “Magnificent 7”
    If Amazon, Alphabet, Microsoft et al do another round of thousands of tech layoffs, and don’t replace those jobs, the trend is obvious.

  3. VC funding for startups steadily declines, especially seed and Series A
    If optimism has vanished, VCs will stop cutting checks; if that number drops below $250B/year, the game is up for “founders creating new work.”

  4. Enrollment for CS at major universities trends down for two consecutive years
    Forget polls and vibes: the 19-year-old signals with her application choice.

Dystopian Coda

If these things happen, welcome to the “salaried gig economy,” in which pushing pixels is piecework, and people who once regarded themselves as high-skill workers enjoy neither the status nor the compensation of the past. And productivity? Sure, it increases, but for whom?

The Progress Scenario: AI as a Force Multiplier, Not an Axe

Now for the case that makes people roll their eyes—because it sounds a bit like every previous era’s optimism about automation, but that’s only because sometimes, things do get better.

The hypothesis: The same dynamic that made accounting software create more accounting jobs, not fewer, plays out again, at a higher intensity. As AI augments routine cognitive work, more White Collar Work happens, not less. Not only does no one “run out of work,” but a whole tier of new jobs, companies, and workflows emerge.

How Do We Get There?

  • The “playing field” is bigger than anyone expects
    Yes, LLMs write code and sales emails. But remarkably, they also dramatically lower the friction for starting something new. Instead of waiting for IT to get to your ticket, you duct-tape something together with GPT-5 and some Zapier scripts. And suddenly, you need an engineer, a marketer, a compliance lead—because your “toy” became a real business in 6 weeks.

  • Small companies start new things faster
    The “two engineers and a dog” cliché becomes “a founder, an LLM, and a dog.” But it’s not a replacement; it’s a springboard. As more people start stuff, they hit ceilings: you need someone really good at marketing, fundraising, design. Instead of fewer jobs, you get more—like the way spreadsheets created more need for MBAs, not less.

  • AI upskills, not de-skills
    The effective “minimum competence bar” rises; more people can tinker, build and sell, raising the tide for early-stage business creation.
    It won’t be frictionless utopia: friction just moves up the value chain.

  • The pace actually accelerates in Year 2
    Here’s the potential moment of surprise: the ability for AI to self-improve (train, evaluate, iterate) creates a feedback loop. We might see a sharp upward kink in the productivity curve—especially for workflows built around AI writing, reviewing, and iterating on code and text.

Falsifiable Markers of the Progress Slide

To know if optimism wins:

  1. Tech job postings recover (and sustain) above baseline—let’s say “Fred Indeed Software Job Postings” read 100+
    If demand for programmers returns to 2021-level, the apocalypse is not arriving.

  2. The “Mag 7” are hiring, not firing
    This is not anecdotal; it’s tracked and public. If Amazon, Microsoft, Alphabet are adding engineers for two cycles straight, it’s a vote for the “AI drives growth, not contraction” story.

  3. VC funding not just recovers but accelerates—$400B+ and >6,000 deals/year
    Let the “seed is the new series A” begin. When money is flowing, the future has believers.

  4. Computer Science majors rise
    If students are running toward the field, not away, that’s your lagging but reliable signal.

Utopian Coda

If these indicators align, it means AI will look more like the spreadsheet or the database than the assembly line robot. Yes, tasks change, get automated, get weird new job titles (“Prompt Engineer,” “AI Flow Designer”), but work creates more work, and technical expertise continues to compound its value.

Why These Predictions Matter

The most important thing about the above predictions isn’t their detail—it’s their falsifiability. At this time in AI history, almost everything said about "job loss" or "job gain" is an unfalsifiable statement, or a vibe masquerading as a thesis. But the data will speak first in the labor market, in funding, and in what ambitious 19-year-olds decide to do with their lives.

We won’t have to wonder which story we’re living. Over the next year and the next two, these indicators will start blinking, one way or the other. If nothing changes—job listings stay anemic, funding clears out, CS enrollments drop—panic. If, instead, more companies are built and the labor market stays tight for engineers, maybe start relaxing.

Either way, you’ll know which world you’re in—faster than you think.