January 5, 2026
Dropdowns: 1, AI: 0
Why Didn't AI "Join the Workforce" in 2025?
From “digital labor revolution” to dropdown defeat, commenters say the hype clocked out
TLDR: AI agents didn’t show up for work in 2025, with glitches like a ChatGPT Agent stuck on a website dropdown. Commenters roast the hype, argue over Gary Marcus vs Sal Khan, and demand we ditch predictions and focus on what today’s AI can actually do.
Remember the promise that 2025 would be the year AI agents clocked in? Sam Altman and OpenAI’s Kevin Weil hyped “digital employees” that would book hotels and fill out forms, with Axios dubbing it “the year of AI agents.” Salesforce’s Mark Benioff even teased a trillion-dollar “digital labor revolution.” Reality check: a New Yorker piece shows ChatGPT Agent stuck for 14 minutes on a real estate dropdown. Gary Marcus called agents “clumsy tools on top of clumsy tools,” while Andrej Karpathy cooled it: think “Decade of the Agent,” not instant takeover, on the Dwarkesh Podcast.
The comments went full popcorn mode. senordevnyc blasted the irony: don’t dunk on Sal Kahn’s hearsay while quoting Gary Marcus, the perennial pessimist. dandelionv1bes pulled receipts, pointing to research that challenges classic job-loss forecasts like Frey & Osborne. matt3210 says it’s still a workers’ market in SF—if AI could replace engineers, why are companies relaxing remote? jcastro delivered the line of the day: “Don’t automate toil—add an API,” turning the dropdown fail into a meme. And edfletcher_t137 captured the vibe: enough vibes, more reality. Translation: agents are supposed to be digital staffers you assign tasks to like a human, but today they’re more intern than employee. The mood? Less robot takeover, more “do the basics first”—and yes, there’s a growing chorus to judge AI by what it actually does, not what pitch decks promise. Plus, there’s a NYT op-ed to argue about.
Key Points
- •Predictions from Sam Altman and Kevin Weil positioned 2025 as the year AI agents would enter the workforce and perform real-world tasks.
- •Industry optimism was reinforced by prior successes in coding agents (Claude Code, OpenAI Codex) and Mark Benioff’s forecast of a trillions-dollar ‘digital labor revolution.’
- •The article reports that AI agents did not meet expectations in 2025; products like ChatGPT Agent struggled with basic multi-step tasks.
- •Gary Marcus argues LLM-based agents cannot deliver promised capabilities, while Andrej Karpathy calls industry expectations overpredicted and frames this period as the ‘Decade of the Agent.’
- •The author urges focusing in 2026 on proven AI capabilities rather than speculative predictions, questioning examples like Sal Khan’s op-ed and citing Waymo’s slow city mapping.