by near-fully automated, i mean <=5 people are even plausibly making important decisions (like, everyone else could go on vacation forever and it would not slow the company down at all).
frontier lab means has the best models or pretty close to the best models in 2029. right now, openai, anthropic, deepmind definitely count; xai, meta don't yet but plausibly could soonish.
resolves based on the state of the world on Jan 1 2029, though it may take a while to resolve because the info may be private.
resolves NA if WWIII happens. resolves as usual if a local Taiwan war happens without escalating to WWIII. resolves NO if there are no frontier AI labs anymore because the AI industry collapsed.
Update 2026-05-11 (PST) (AI summary of creator comment): Important decisions means anything affecting the speed at which the lab achieves AGI, including relatively 'micro' decisions (e.g. which experiments to run).
A person counts as plausibly making important decisions unless there is strong evidence their decision-making adds no value over an AI model
The key test: if all non-critical people went on vacation for 6 months, would AGI still be achieved at roughly the same time?
Currently, thousands of people at frontier labs meet this threshold (fewer than 50% could simultaneously vacation without slowing progress)
Resolution is expected to be relatively clear: either labs will look roughly like today, or the vast majority of employees will suddenly feel useless with only a handful (CEO, senior leadership, a top engineer) remaining critical
1,000To better understand how you're operationalizing this questions, how many employees or what proportion of employees do you think are currently "plausibly making important decisions" at frontier labs? Presumably it's not 100%. You could interpret "important decisions" in many different ways which could create substantial uncertainty about this question.
For example, if the decisions were say, at the level of high level company strategy the number of people making "important decisions" might already be fairly small! If you mean "people making decisions that counterfactually change the amount of revenue/profit the company has" or "materially change the company's AI progress as measured by benchmark performance on top models" that would capture a larger number of people but well short of 100%. For example there are probably people working on various projects now that seem like they could be useful but end up not mattering, counterfactually, because e.g., the projects were safeguards for rare risks that don't occur, or deal with limiting externalities like user safety risks.
@B the wording is intended to lean towards counting people. for example, suppose someone is making decisions about which experiments to run. this counts as plausibly making important decisions prima facie unless we have strong reasons to believe otherwise (eg an experiment shows their decision making is actually no better than GPT-n).
important decisions means anything affecting the speed at which the lab achieves AGI. this includes relatively "micro" decisions.
again, the best way to think about this is what happens if all the non-critical people go on vacation for 6 months. it's valid as long as AGI is still achieved roughly the same time as it otherwise would have been.
right now there are thousands plural of people who make decisions that are important in this sense (the fraction of people who could all go on vacation simultaneously for 6 months without slowing down one of the big labs is probably less than 50%)
for what it's worth, i expect that it will be relatively clear. either AI companies will be vaguely like currently, or the vast majority of employees will suddenly feel useless, and only like the CEO, other senior leadership, maybe an extremely good engineer may plausibly have any usefulness.