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

Ben Todd: Careers as the “biggest lever” for the AI future

Ben Todd (co-founder of 80,000 Hours) argues that careers are a person’s “biggest lever” for shaping the AI future. He also contends that most people can still have outsized impact even if they aren’t working directly in AI today—by choosing careers that maximize positive effect during rapidly changing AI timelines.

AI timelines: plan around your peak impact

Todd says career planning debates often get stuck on when AGI/superintelligence arrives. Instead, he emphasizes choosing careers based on when your personal impact will be greatest under different plausible scenarios.

He describes three broad acceleration possibilities:

1. Fast takeoff AI automates parts of AI R&D quickly, creating feedback loops. Powerful autonomous systems arrive before many jobs or robotics are automated.

2. Medium timeline Similar progress occurs, but without the same rapid acceleration—more constrained by compute scaling and other bottlenecks.

3. Longer plateau (least likely) Progress slows if current paradigms become uneconomical or if compute scaling hits a wall.

Across all but the most extreme rapid-takeoff views, Todd argues there is still time to position yourself well.

The 80,000 Hours career framework for impact

Todd reiterates a five-part “career option” framework:

• Impact
• Career capital (skills, credentials, connections, character-fit, etc.)
• Personal fit
• Exploration value (learning what works best for you)
• Other factors, folded into fit/satisfaction goals

He notes that short time horizons (when key opportunities are near) reduce the emphasis on exploration/career capital, but don’t eliminate them. Skill-building still pays off, even over a few years, if it meaningfully increases productivity.

He also reframes “timeline” away from AGI dates to the window when your actions matter most.

What to work on: “big, neglected, potentially solvable” problems

Todd applies 80,000 Hours’ problem-selection lens (big / neglected / potentially solvable) to AI safety. His top priorities include:

1) Loss of control of autonomous AI

• Potentially irreversible outcomes like human disempowerment
• Still under-resourced relative to broader AI capability efforts (often thousands focused on safety vs. 100k–1M potentially focused on capability acceleration)

2) Concentration of power

• Scenarios where one actor gains a huge “digital workforce” by accelerating capabilities—potentially widening gaps under superexponential growth
• AI-enabled surveillance and centralized control (e.g., governments monitoring more effectively than before)
• Limited practical attention to centralized-instructions and governance mechanics

3) Engineered pandemics

• AI may lower barriers to designing and releasing high-impact pathogens
• Even if harder than nuclear weapons, it could be easier to execute and may still spread via lab accidents

Upside vs downside: why safety work is still central

Todd argues that focusing on downside risks can have unusually high value because it helps prevent scenarios where the “best future” never arrives.

He contrasts this with simply speeding up progress: if the world is already racing toward better outcomes, speeding the timeline yields less benefit than preventing existential catastrophe.

He frames safety as an impact portfolio problem, not a single-bet strategy.

Where to work: frontier labs vs outside options (and “question your motives”)

Todd says the choice between frontier companies and external organizations is complicated.

• Pros of working in frontier labs

  • Access to the best technical teams
  • Ability to implement research directly
  • Potential to catch qualitative shifts in alignment/control behavior between model generations

• Pros of working outside labs

  • Safety research can still be effective (e.g., groups like Redwood Research)
  • May preserve independence

A repeated central theme is motive checking: people should ask whether they’re choosing a lab role for safety reasons or for status/coolness/convenience.

If someone does go inside a lab, Todd suggests assessing whether they can remain aligned with their values under pressure. He proposes practical self-assessment methods, such as:

• Objective evaluation of a company’s track record
• Trusted friends to call you out
• Written pre-commitments
• Choosing a culture rather than relying on personal resistance later

AI safety “accelerant” critique: portfolio answer

Todd acknowledges the critique that safety work can accelerate capability development (e.g., RLHF improving usability and adoption).

Still, he argues the better counterfactual is having a thriving safety ecosystem that reduces major risks, rather than relying on “no safety work” scenarios.

He supports:

• Pause advocacy (possibly a temporary pause near certain capability thresholds)
• While simultaneously funding work on “no pause” scenarios

Again, this is framed as portfolio thinking.

Policy: what matters and how it becomes enforceable

For AI policy, Todd emphasizes building the prerequisites for enforceable regimes, including:

• Compute tracking / the ability to monitor major training runs
• Capacity for an “off switch” (rapid compute shutdown in replication-like scenarios)
• Transparency about capability progress
• Red lines / emergency response plans for dangerous behavior

He also argues that Overton-window progress depends on political will and AI literacy in government, not just having good proposals.

Pandemic prevention: use AI skill for bio-risk “startup-like” execution

Todd suggests concrete pandemic-related work that intersects with AI capabilities, including:

• Improved wastewater monitoring to detect outbreaks early (including exponential-growth signals)
• Red teaming gene synthesis providers to reduce misuse risk
• Better PPE, air filtration, and stockpiles for severe pandemics
• Rapid vaccine development

He describes these as engineering-and-execution opportunities where AI-capable builders could reduce worst-case outcomes.

Joining vs founding nonprofits/organizations

Todd says there is plenty of funding in the nonprofit AI risk space and that both approaches can work:

• Joining existing organizations can leverage established scaling and execution
• Founding can fill real gaps, but it’s hard and not for most people—entrepreneurial founders may overvalue the satisfaction of creating something new

He also mentions programs/networks that help career transitions and organization-building, such as:

• Successive
• Horizon Fellowship transitions
• Halcyon
• Catalyze Impact
• 6 Impact

“Newer / neglected” topics: digital minds and space governance

Todd flags speculative but potentially important areas that may be neglected when attention shifts:

• Digital minds / moral standing / consciousness Prepare for future debates if AIs resemble humans closely enough to raise rights-like questions.

• Space governance If AI accelerates space expansion via self-replicating probes, first-mover dynamics could trigger a land-grab. He argues for better modeling and a dedicated governance research institute.

He also mentions gradual disempowerment—even with aligned AI, humans may be economically sidelined.

Positive-future steering: “viatopia” over utopias

On using fiction/narratives to shape values, Todd is skeptical of “utopian” writing due to poor track records. He endorses Viatopia (from Will MacAskill):

• Focus on robust navigation goals
  • avoid irreversible existential risk
  • avoid authoritarian lock-in
  • improve information quality
• without specifying a fully detailed utopia

Broader motivation: applying AI skills beyond AI risk

Todd argues that AI skills—especially coordinating AI agents to do real work—transfer broadly and can help many causes.

He also notes:

• Poverty reduction could benefit strongly in a world where AI makes many services cheap and scalable
• Preventing exclusion from the AI “windfall” matters, tied to concentration-of-power concerns
• He floats the idea of a grand bargain among major countries to reduce racing dynamics in exchange for shared benefits

Presenters / contributors

• Ben Todd — co-founder of 80,000 Hours; author
• Cognitive Revolution host — unnamed in the provided subtitles (interviewer)