OpenAI CEO Sam Altman: Good Founders Never Run Out of Ideas
Sam Altman stated that you can give a founder an idea to start a company, but the issue is that they must come up with new ideas, crazy ideas, and major pivots almost every week, continuously.
He recalled an experiment at Y Combinator: funding 20 founder teams that were strong in other aspects but lacked ideas, all of which failed. The conclusion is that good founders are those who consistently have ideas.
He broke down this ability into intelligence, creativity, and independent thinking, with the core being an intelligence that can view problems from different angles and conceive things that do not yet exist but should. Without this, the role of a founder does not exist.
He wrote more firmly in the Startup Playbook: ideas should ideally precede the company; once a formal company is established and direction is rushed to be found, ideas tend to be less crazy, ultimately becoming reasonable-sounding but derivative solutions, which is the danger of pivoting. Good founders usually have too many ideas rather than too few.
In the same logic, he also wrote: stuffing ideas into founders almost always fails; those who want to start a business but cannot come up with a topic should first practice generating ideas, rather than registering a company first. The environment should be close to those who have a sense of the future, are willing to discuss outrageous plans, are optimistic, creatively smart, have high idea flux, and are not ashamed of silly ideas. The best ideas often sound silly at first, making them fragile.
In market mechanisms, buyers are accelerators and early-stage funds, while sellers are "people who can continuously generate ideas" rather than "those who can execute given topics." The event was driven by YC's real-money controlled experiment: execution can be packaged and sold, while creativity cannot be outsourced, making the combination invalid. Funding flows to those with high idea flux; benefiting are founders who can turn the first seed idea into new decisions every week; under pressure are career manager-type entrepreneurs with impressive resumes who need others to provide topics. Once a company is established, tolerance for crazy ideas decreases, and derivative ideas begin to dominate boardroom discussions.
Source: Public Information
ABAB AI Insight
Altman himself was one of the first founders at YC, created Loopt, and later managed YC before founding OpenAI. He has repeatedly navigated both paths of "finding people after having an idea" and "finding an idea after having people": Reddit was rejected when Paul Graham accepted it, making it one of the few exceptions where the founder survived by changing the topic after passing the initial test; the 20 teams without topics during his tenure were a controlled failure. Loopt itself did not grow into a giant, but he wrote the "weekly new idea" as a filter to explain why execution can be outsourced, but idea generation cannot. OpenAI's transition from a nonprofit research lab to a product company involved multiple narrative changes, governance adjustments, and product form changes, embodying the continuous topic changes he described.
The capital path is: seed checks do not buy business plans, but rather a person who will continue to produce a batch of plans even after the initial plan becomes obsolete. About one-quarter of YC batches appear to have the potential to grow into billion-dollar companies, but only a few actually do; he has observed that after a batch ends, the growth curve becomes steppy and flattens, attributed to momentum loss and idea exhaustion. Money flows from LPs to accelerators and then to founders, with the truly priced scarce resource being the perception of "things that do not yet exist but should," rather than the ability to execute existing solutions. Derivative ideas are easier to finance because they sound like validated markets, but this is precisely the trap that the 20 teams fell into in the experiment.
Analogous to Paul Graham's "living is better than dead," Peter Thiel's "secrets," and music labels giving producers a style but finding that artists who cannot write choruses cannot produce albums. In terms of industry positioning, accelerators have shifted from the early dogma of "people over ideas" to a narrower correction: people must bring their own idea flux, or they are not enough. The current stage is about controlling selection criteria, not expanding batch sizes.
Structural judgment belongs to the transfer of pricing power. The mechanism is: a startup's product is no longer the same product after the first week as it was on opening day; pricing power belongs to those who can continuously redefine it. Providing a business plan along with an execution team buys the first version; the market wants to buy the generator from the 2nd to the 200th version. Without a generator, the check becomes a prepayment for a document that will soon expire.
ABAB News · Cognitive Laws
Those who can execute given topics cannot create companies that must change topics weekly.
Having a company first and then finding ideas means ideas cannot be too crazy.
Idea flux is the layer of intelligence that founders cannot outsource.