The last 300 days of work? (No, but...)

TL;DR

• 1

  The "last 300 days of work" rumor is real, but thinly sourced: Shapiro traces it to Kevin Roose's tweet about overhearing the phrase at an AI lab, and treats it as plausible signal rather than proof of imminent labor collapse.

• 2

  Frontier labs may be seeing capability saturation before the real economy does: He points to benchmarks like GDPval and heavy agent spending at top labs, where companies are burning millions of dollars a month on tokens to learn how to build better agents.

• 3

  AI is already swallowing grunt work in small companies: Shapiro's wife now does much of her marketing, social, and customer research work through Claude, Gemini, Canva, Slack, and other products that now ship with built-in copilots.

• 4

  The real near-term effect is slower hiring, not instant layoffs: He cites reporting that new grad hiring is down 9 to 13 percent at companies using generative AI more heavily, with new grad unemployment around 5.6 percent versus a broader rate under 4.8 percent.

• 5

  Corporate inertia is the biggest brake on AI replacing work: Drawing on experience at Cisco and Advance Auto Parts, he argues that Fortune 500 planning cycles, legacy systems, security reviews, and "if it ain't broke, don't fix it" thinking push large-scale change far beyond 300 days.

• 6

  Fully digital jobs are much more exposed than physical or infrastructure-heavy work: Shapiro says agentic AI could probably do most of his old KVM-style infrastructure role within 300 days, but not the broader stack that still depends on power, cables, hardware, backups, and identity management.