Summary: Fortune published Matt Shumer’s essay “Something big is happening in AI” on February 11, 2026; Fortune says it exceeded 50 million views. Shumer, OthersideAI’s cofounder and CEO, argues paid AI capabilities outpace free tiers by more than a year and describes agents that generate, test, and repair code. Critics questioned the one-to-five-year disruption forecast and his conflict of interest as an AI company leader.
“Something big is happening in AI” reached Fortune’s pages on February 11, 2026, written by Matt Shumer, cofounder and CEO of OthersideAI. His personal site and X carried the piece first, and by Fortune’s count more than 50 million people have viewed it. A long-form essay resting on argument alone — nothing in the news cycle to tie it to — ended up lighting the defining AI discourse fight of early 2026. Worth unpacking are the claims themselves, the evidence behind them, and the shape of the response.
A COVID Metaphor, 50 Million Views, and a Verdict
COVID anchors the entire essay. Dial back to February 2020 and the pattern is familiar: a few people paying close attention while nearly everyone else filed the warnings under overreaction. That, in Shumer’s telling, is where we now sit — the “this seems overblown” phase of something “much, much bigger than Covid.” The verdict arrives without hedging: the “AI is hitting a wall” debate is over; in the medium term, “nothing that can be done on a computer is safe”; and law, finance, and medicine face disruption on a one-to-five-year clock, not a ten-year one.
Free Tier Lags the Paid One by Over a Year
A structural observation is what pushed the essay beyond tech circles to a general audience: paying users have access to capabilities that the free version trails by more than a year. In his words, “Judging AI based on free-tier ChatGPT is like evaluating the state of smartphones by using a flip phone.” The mental model most people hold froze on a model more than a year old, while $20 a month buys agents that run entire workflows end to end. His picks for the best models at the moment: GPT-5.2 and Claude Opus 4.6.
Self-Testing Agents: The Evidence and Its Ceiling
His own daily work supplies the technical evidence. The routine: describe the desired outcome in plain English, step away for four hours, and return to finished work — better, by his account, than he would have done himself. A Monday anecdote adds specificity: the AI wrote tens of thousands of lines of code, then opened the app itself, clicked through the features, tested, iterated — “like a developer would” — and delivered the final report: “It’s ready for you to test.”
The detail an agent engineer should fixate on isn’t volume; it’s the closed verification loop: generate, operate the UI, test, fix, repeat. That loop is the mainstream 2026 coding-agent playbook, and it is the same agentic engineering capability Zhipu put front and center when open-sourcing GLM-5 that same week. Shumer’s departure is the forecast attached: he believes the loop is about to swallow every job that runs on a computer.
Jobs on a One-to-Five-Year Clock
The prediction he leans on comes from Dario Amodei: AI will eliminate 50% of entry-level white-collar jobs within one to five years — and Shumer notes that many people now consider that figure conservative. For backdrop, February 2026 had watched software stocks shed roughly $1 trillion in value over a single week. The advice to readers stays practical: push AI into real work instead of parking it at a search box, run an hour of experiments a day — enough, by his estimate, to put you ahead of 99% of people — build savings, and lean into work that resists replacement: trust, physical presence, licensed accountability. The essay closes on a line that has been widely quoted since: “The future is already here. It just hasn’t knocked on your door yet.”
The Counterblow: Conflict of Interest and Alleged Hype
If the spread was loud, the backlash matched it. Cato Institute ran a response under the title “…but That’s the Only Thing Matt Shumer Got Right”: it concedes that AI can complete time-intensive, complex tasks at record speeds and rejects the storm-is-coming, hunker-down framing. Forbes arrived two days later with a critique whose target was the virality of these manifestos itself, and across press coverage Gary Marcus wrote the essay off as alarmist hype. The criticism raised most often is the conflict of interest: the voice declaring that AI is about to upend everything belongs to someone who runs an AI company.
Look closer and the two camps share one fact set; the dispute is over timelines and who absorbs the distributional consequences. For product teams, the defensible ground sits in the middle: the free-versus-paid gap is real, frontier agent capability is real, but “everything flips in one to five years” is an unfalsifiable forecast. The parts that can actually be verified — your own workflows, your own evals — are testable today.