AI AGENTS DEBATE: These Jobs Won't Exist In 24 Months!

Unlike ChatGPT’s request–response mode, AI agents accept a goal and keep working—sometimes for 30+ minutes—until they either achieve it or fail and ask for help. They can be given tools such as browsers, coding environments, payment methods, and APIs. This already enables tasks like ordering goods end‑to‑end or building and deploying full software applications from natural language alone, and coherence over longer horizons is improving exponentially.

Any job that is “as routine as it comes”—QA, data entry, many customer support roles, basic accounting and legal drafting, parts of radiology and medical imaging, and other text‑in/text‑out or repeatable click‑work—is directly exposed. Klarna already replaced the equivalent of 700 support agents with AI; Replit cut ~70% of support load. High‑status knowledge jobs such as anesthesiologists are also vulnerable once AI can monitor, recommend, and titrate with high precision, leaving a single human supervising many wards.

The panelists frame the future as favoring “high‑agency generalists” who can spot opportunities, design systems, and coordinate many AI agents rather than do routine execution. One person can now build a SaaS business in minutes, run M&A processes with a fraction of legal costs, or generate internal tools for HR or finance on demand. Small teams of 5–10 people armed with AI can create more in 3 years than many large organizations did in decades—if they move early and learn to use the tools deeply.

AI democratizes capability—anyone with ideas and an internet connection can build—but variance in outcomes spikes. The best “AI‑leveraged” entrepreneurs might become 1,000× more effective than average. Distribution, audience, and authenticity become key moats once capital and coding are commoditized. Meanwhile, elites can use AI for surveillance, financial manipulation, and even physical coercion (e.g., autonomous drones), risking a scenario with ultra‑wealthy “owners” and many economically sidelined ‘useless eaters’.

Already, deepfake ads and voice clones are defrauding fans of the host and guests. In the near future, undetectable video forgeries, personalized social‑engineering bots, and AI‑driven cyberweapons could destabilize trust, politics, and economies. At the same time, platforms are strongly incentivized to deploy AI defenses—scam detection, content authenticity checks, and personal “AI antivirus” agents. A cat‑and‑mouse dynamic between offensive and defensive AI is inevitable, with no guarantee that defenders win.

Static skills are quickly automated. The panel argues for cultivating general reasoning, adaptability, and “navigation” in complex systems: prototyping, iterating, and reading feedback instead of memorizing blueprints. One‑to‑one tutoring shows the strongest educational impact, and AI can finally provide scalable personalized instruction to every child. For their own kids, they stress chess, jiu‑jitsu, building things, entrepreneurship (e.g., lemonade stands), and using AI creatively while limiting addictive consumption.

Even before AI, rapid change has outpaced our ability to adapt, feeding loneliness, mental illness, and falling birth rates. AI risks intensifying this: endless entertainment, bespoke pornography, and perfect virtual worlds could erode motivation for real relationships, struggle, and parenthood. UBI alone is unlikely to provide meaning; humans need to feel useful. Without deliberate design of new “meaning ecosystems”—real projects, communities, and rites of passage—AI abundance could leave billions psychologically adrift.