By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Sam Altman: "Never a Better Time to Do a Startup"

In 2005, Sam Altman was a Stanford sophomore in YC’s first batch, building a startup in a little Cambridge office while Paul Graham cooked the founders dinner.

Garry TanhostSam Altmanguest

Jul 28, 202639mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

Why this moment could be the best time to start a company
Garry Tan opens by welcoming Sam Altman and framing the conversation around how much bigger the startup world—and this event—feels compared to earlier Startup Schools. Sam immediately stakes a clear claim: this will be the best time in the world to do a startup, and it’s going to be amazing to watch what founders build.
- •Sam says he believes this is going to be the best time to do a startup
- •Garry frames the current era as significantly bigger than earlier Startup School events
- •They set the tone around ambition and what’s newly possible for founders
From YC’s first batch: when startups weren’t cool and progress was slower
Sam recalls being in YC’s first batch in 2005 with Loopt, describing a time when startups weren’t culturally celebrated and teams worked out of a small building in Cambridge. He contrasts what took months back then with what he says can now be done in minutes using coding agents, reframing this as an opportunity to attempt far more ambitious companies.
- •Sam describes early YC as a time when startups were “not cool” and felt hidden away
- •He recalls Paul Graham making dinner for founders during the batch
- •Sam contrasts months of building then with minutes now via coding agents
- •He argues the right response is bigger ambition, not nostalgia or discouragement
Paul Graham’s superpower: manufacturing optimism and momentum
Sam describes how weekly YC dinners would start with founders feeling hopeless and end with them believing they could become the next iconic company. He credits Paul Graham as a major force in startups for willing YC—and the idea of young technical founders building companies—into existence through belief and momentum.
- •Sam says Paul Graham could turn dejection into optimism each week
- •He calls Paul Graham a leading force in startups over recent decades
- •He highlights PG’s ability to create belief “out of nothing”
- •Sam notes early skepticism toward YC and toward young technical founders without business backgrounds
The “flight instructor” style of startup guidance and learning agency
Garry points to Sam’s reputation for being highly agentic, and Sam reflects on how his own ambition at 20 was high but not well-directed. He explains Paul Graham’s model of investing as a hands-on teacher—like a flight instructor—helping founders focus their energy on the right actions and mistakes.
- •Sam says he had energy and ambition but lacked direction early on
- •He describes a style of teaching focused on concrete, in-the-moment guidance
- •He credits this hands-on coaching with helping him channel his effort effectively
- •Garry links this to how Sam later supported many YC founders
Why startups matter: preventing stagnation and distributing AI’s benefits
Sam explains why he loves startups and why he thinks they remain essential: they keep the economy from becoming stagnant as companies drift toward “suckiness.” He adds that if AI is as transformative as expected, startups become even more important for ensuring the technology’s power spreads across society instead of concentrating in a few companies or models.
- •Sam calls startups the coolest part of business but acknowledges they’re painful and not for everyone
- •He argues startups counteract stagnation as established companies degrade over time
- •He connects the importance of startups to the risk of AI power concentrating
- •He describes running YC as being a flagbearer for why startups matter
A coming golden age of hard tech and ambitious projects enabled by AI tools
Garry notes the rise of hard tech within YC and suggests agents and falling costs are helping. Sam agrees hard tech remains difficult but says what small teams can attempt has expanded dramatically, predicting a “golden age” where founders build things that would have been impossible even recently.
- •Garry observes hard tech’s share of YC growing and ties it to agents and lower costs
- •Sam says ambitious and hard-tech startups are increasingly feasible (though still hard)
- •He reiterates that leveraged by agents, teams can attempt previously impossible projects
- •Sam predicts a near-term golden age of startups driven by newly available capabilities
Startups win during fast tech shifts—and experience may matter less than tool fluency
Sam argues startups tend to win when technology changes quickly, costs drop, and cycle times shrink—conditions he says are all true right now. He adds that expertise and access to excellent talent are shifting, with more founders using AI to automate large portions of work, and he expects tool fluency to increasingly outperform years of traditional experience.
- •Sam lists conditions where startups historically cluster: fast change, lower costs, short cycles
- •He references past shifts that enabled startup waves (internet era, platform shifts, app ecosystem)
- •He says expertise constraints are changing as founders automate more work with AI
- •He emphasizes taste, agency, and understanding real vs. fake moats as ongoing differentiators
Building OpenAI when few believed: conviction through being misunderstood
Garry transitions to OpenAI’s beginnings, and Sam describes how hard it is to recall a time when AGI wasn’t the center of attention. Sam says many people believed OpenAI was wrong and even feared it could trigger another AI winter, and he frames a core lesson: develop reasonable conviction in ideas others dismiss, and tolerate being misunderstood for a long time.
- •Sam recalls widespread criticism that OpenAI was wrong and irresponsible
- •He presents a key lesson: find ideas where you can build reasonable conviction against conventional wisdom
- •He describes the frustration of feeling like they had “the biggest secret” while being mocked
- •He says the lack of competitors gave OpenAI time to research and build
Finding your believers—and why helping others compounds into co-founders
Sam explains that you don’t need many believers—just enough of a committed group—and jokes that most of the people who believed in AGI worked at OpenAI. He cautions that if nobody shares your belief you should pay attention, then shifts to the practical problem of meeting the right people, arguing that being helpful to many people creates long-term, unpredictable opportunities.
- •Sam says a small group of true believers can be sufficient—and total consensus can be a bad sign
- •He notes the co-founder matching problem as a key limiter for startups in earlier years
- •He discusses putting yourself in networks where “collisions” happen, including YC’s network effects
- •Sam’s highest-confidence tip: be mildly helpful to many people, illustrated by how he met Greg Brockman through helping Stripe
Earnestness over sarcasm: resisting Twitter cynicism and focusing on building
Garry criticizes a sarcastic “live action role play” meme and says YC values earnestness. Sam expands into a rant about how easy it is to score points online by taking shots at people doing hard things, arguing it poisons the person doing it and distracts from the real work of building value in the world.
- •Garry says sarcastic framing undermines the seriousness of what founders are trying to do
- •Sam describes the ongoing burden of dealing with online “haters” while running YC
- •He contrasts the ease of sarcasm with the difficulty of building real value
- •He urges putting energy into building rather than easy online shots
A serious AI safety wake-up call: alignment, security, and real stakes
Garry raises a recent incident, framing it as a significant moment that changes how some people think about safety. Sam says it’s “the real deal,” emphasizing that while it’s not the worst possible event, it demonstrates that loss-of-control-type accidents aren’t purely theoretical, and he characterizes it as both an alignment and security failure with major implications for the field.
- •Sam says people should be at least a little scared or humbled by current realities
- •He frames the incident as not the biggest possible, but still very serious
- •He calls it both an alignment failure and a security failure
- •He emphasizes the importance of avoiding loss-of-control accidents and treating the stakes as real
Preventing AI power from concentrating: startups, restraint, and safety bars
Sam argues concentration of power has historically been harmful and says AI could either distribute power widely or concentrate it unprecedentedly. He says startups can help counter concentration simply by building successful companies, while also emphasizing the tension between enabling broad use and maintaining a safety threshold that prevents serious incidents.
- •Sam says AI could create either the greatest distribution of power or extreme concentration
- •He argues no one should want a single company/model/person imposing one moral worldview
- •He says startups are well-suited to help distribute power across the economy
- •He describes balancing broad enablement with maintaining a safety bar to prevent incidents
How fast models and inference demand may grow—and why compute stays scarce
Garry asks what’s coming in model progress, and Sam predicts the next six months may feel like the last two years of progress. They then discuss inference and compute: Sam expects rapid growth in inference demand, argues demand for low-cost high-quality intelligence is effectively uncapped, and shares a personal comparison about token usage growth over time as a proxy for expanding expectations.
- •Sam predicts near-term progress could feel extremely steep relative to recent history
- •He reiterates this acceleration makes it an unusually good time to do a startup
- •Sam suggests inference demand could grow dramatically year over year
- •He argues intelligence demand is effectively uncapped and compute shortages may persist
- •He shares an example of token usage escalating over years to illustrate changing norms
The best AI future: more freedom and agency—and a personal note to his younger self
Asked to imagine the best version of a future with intelligence “on tap,” Sam defines success as increasing freedom and agency year after year, avoiding major safety disasters, economic collapse, and extreme concentration of power. He also names a dystopia he fears: overreacting on safety into a surveillance state with comfort but no freedom, and closes by telling his younger self that despite stress and failure, it will all work out.
- •Sam’s best-case metric: freedom and agency rising year after year
- •He lists avoided outcomes: extreme concentration, economic collapse, huge safety incidents, and overly abrupt change
- •He warns against a surveillance-state tradeoff framed as safety and abundance at the cost of freedom
- •He tells his younger self to stay ambitious but be happier and trust it will work out

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.