Your Biggest Lever: Designing your AI Career for Maximum Impact, with 80,000 Hours founder Ben Todd

What It Covers

Ben Todd, cofounder of 80,000 Hours, discusses how individuals can position their careers for maximum impact during the AI transition. The conversation covers AI timeline planning across three scenarios, the top three global risks (AI control loss, power concentration, engineered pandemics), and concrete career pathways across technical research, policy, communications, and organization building.

Key Questions Answered

• •Career Timeline Reframe: Rather than asking when AGI arrives, ask when your personal impact will peak. Todd recommends planning across three scenarios: fast takeoff by 2027-2028 via automated AI R&D, medium timeline reaching powerful AI by the 2030s, or a prolonged plateau. Even under short timelines, a five-to-ten year planning horizon justifies skill investment — a one-year retraining that yields 20% productivity gains pays off within four to five years.
• •Top Three Priority Problems: Todd ranks loss of control over autonomous AI first, given potentially irreversible human disempowerment. Power concentration ranks second — a single company achieving exponential AI growth could accumulate nation-scale digital workforce power. Engineered pandemics rank third, with AI lowering the barrier for state and non-state actors to create pathogens far deadlier than naturally occurring ones. Roughly 1,000-2,000 people work on these risks versus potentially one million on capabilities.
• •Four High-Leverage Career Categories: Technical research (AI evals, control, interpretability) now skews toward engineering over conceptual work — Metr alone has 20 high-value projects but capacity for only two or three. Government and policy roles need people who bridge technical and governmental worlds. Communications work to raise public understanding remains severely understaffed. Organization building — management, legal, HR, recruiting — is needed across all organizations tackling these risks.
• •Frontier Lab Employment Decision: Working at frontier labs offers unmatched access to frontier models and direct implementation of safety research, but carries the risk of accelerating capabilities. The decision hinges on personal P(doom) estimates and whether alignment research is tractable. Todd advises writing down specific pre-commitments about when you would act or leave, cultivating friends who will call out rationalization, and choosing organizations whose culture aligns with your values from the start.
• •Funding Environment and Org Strategy: The AI safety nonprofit space currently has more funding than talent. Coefficient Giving has been supplemented by new funders, and Anthropic founders have pledged roughly 80% of their equity — potentially tens of billions — to philanthropy. Todd recommends evaluating whether joining a high-performing existing organization and multiplying its effectiveness by even 5% outperforms founding a new one, since entrepreneurial bias systematically overweights the satisfaction of building from scratch.
• •Concrete Policy Priorities: Todd identifies compute tracking infrastructure as the most foundational near-term policy goal — without it, a strategic pause becomes unenforceable. Additional priorities include establishing industry-wide red lines with agreed emergency response triggers, creating government capacity to detect an intelligence explosion within days rather than months, and building the political groundwork now for a potential future administration to implement a strategic pause, including a bilateral deal with China, which has stronger incentives to pause since it is currently behind.
• •Neglected Emerging Areas: Three underexplored problems warrant early attention. Digital minds and AI welfare will become unavoidable as AI systems become behaviorally indistinguishable from humans, and philosophical uncertainty means policy frameworks need development now. Space governance has near-zero institutional attention despite potential first-mover lock-in dynamics once self-replicating AI probes become technically feasible. Gradual human disempowerment — where even aligned AI outcompetes humans economically — lacks any concrete prevention proposal beyond hoping aligned AI advises against it.