Paul Graham, Founder of Y Combinator on Startups, Ambition, and Great Founders

YC batch scale & cadence

• 47th YC batch; YC has been running for 21 years.
• PG notes the YC talk is built from scratch each time, via a quick text-file process rather than reusing a fixed script.
• Core idea: “Most of starting a startup is always the same”—technology changes, but execution patterns persist.

How YC is “more serious” now (vs. old claims)

• PG challenges the narrative that YC “jumped the shark.”
• Critics often argue YC was better “in the old days”, implying it’s worse now.
• PG’s counterpoint: the earlier “old days” examples weren’t truly world-scale ambition like the bets seen today.

Examples of high-ambition bets (serious startup themes)

• Intercontinental ballistic cargo

  • A joke analogy used to contrast seriousness vs “lighter” earlier internet outcomes.

• Curing cancer

  • PG cites multiple approaches (e.g., vaccines vs therapies).
  • He mentions a YC-batch startup doing on-demand research for cancer, framed as “death of a thousand cuts” rather than a single “silver-bullet cure.”
  • He ties this to the YC community “flywheel”:
    • The company and/or its people helped Sid (GitLab founder) during his cancer.
    • PG references Sid’s founder-mode mindset essay.
  • Actionable implication: winning companies may emerge from repeatable playbooks—e.g., “what Sid did for himself” becomes a broader product.

Founder ambition as a core operating driver

• Implied framework:
  • Founders must be ambitious, but the day-to-day engine is often fear of disaster/failure, not just eventual wealth.
• PG’s framing of “ambition”:
  • Technical skill alone isn’t enough; obstacles are “too fearsome.”
  • Even if a founder could become rich, what motivates them moment-to-moment is avoiding near-term catastrophes (e.g., server crashes).
  • When success happens, it can be mathematically surprising:
    • PG notes he did the math for founders becoming “billionaires” based on last-round valuations.

Prestige vs. purpose: why “bad reasons” to do startups fail

• Critique: some founders pursue YC as a credential/badge (like applying to Harvard for prestige).
• PG’s argument: startups have no easy “major.”
  • They force “theoretical physics”: you must be hardworking and determined.
• Takeaway:
  • Startup-building is least efficient for appearing cool.
  • Recognition comes years after brutal execution.

YC language & selection lens: “formidable” founders

• “Formidable” is described as a YC-term (originating in YC founders’ private language).
• Definition / playbook logic:
  • Someone who gets what they want in any situation.
• Alignment mechanism:
  • If investors hold equity, then interests align—if founders get what they want, investors benefit.
• Therefore:
  • Investing in formidable founders improves odds more than chasing a single idea.

Lean startup / parallel work debate

• Asked about Patrick Collison saying “lean startups may be dead.”
• PG response:
  • He isn’t fully aligned with what “lean startup” means specifically (he hasn’t read The Lean Startup).
  • He rejects “lean is dead” because:
    • You can start with little money by scaling scope to milestones.
    • If you can reach a convincing milestone, you can raise the next round.
• Examples:
  • Rocket startup:
    • You can’t build the rocket immediately.
    • You build a design, simulate, and show to experts—then raise funds after credibility.
  • Tokens are expensive due to GPU shortages, but token/inference costs should fall as compute improves (a cost curve that makes execution easier).
• Operating principle:
  • Adjust plans to available budget; optimize for milestones, not initial capital intensity.

AI perspective (business relevance to execution)

• PG’s noted surprise:
  • Expected AI progress from “perfect small” (fly → mouse → cat → human) but instead got “full human but wrong/bullshitting undergraduate” behavior.
• AGI is not a crisp finish line:
  • There’s a “smear”—some capabilities are far beyond, others lag.
• Practical business takeaway:
  • AI doesn’t remove execution needs—startups still require fast iteration.

Core startup KPI: shipping pace

• PG’s explicit metric: pace of shipping new stuff is the best predictor of success.
• AI changes costs/constraints:
  • Old dominant cost: salaries.
  • New dominant variable cost: GPU/compute and token spend (PG highlights tens of thousands of dollars per day in token costs).
• Implied KPI discipline:
  • Measure and manage shipping speed and compute burn (AI bills) to sustain iteration.

How YC provides execution leverage (“YC GDP”)

• YC advantages:
  • Startups are usually lonely.
  • A YC batch creates a collegial environment where peers solve technical problems quickly by asking each other.
• “YC GDP”:
  • Startups can sell products into the batch ecosystem—early adopters who decide quickly and will listen to pitches.
• Go-to-market implication:
  • Use YC batch participants as a rapid customer discovery / early adoption channel.

Origin story of YC (process and iteration)

• YC idea began as:
  • A plan to be an angel firm, investing early with standardized paperwork (smaller amounts earlier).
• PG/YC reframed:
  • They weren’t sure how to be investors, so they would learn by funding many startups at once (batch).
• Batch format origin:
  • Modeled after undergrad summer jobs—batch runs in summer to “replace summer jobs.”
  • PG says the “perfect length” was an accident.
• Operating lesson:
  • Build an operating system (batch investing + structured pipeline) and refine it as you learn.

How YC changes with scale

• PG says YC has changed “very little” in philosophy:
  • More startups exist, but the problems are similar.
• Organizational structure:
  • YC is effectively divided into pods (each startup part of a group),
  • so the experience resembles earlier batches (e.g., 2012, when there were 70 startups).

Where the next trillion-dollar company comes from

• PG’s answer: it comes from the right founders (returning to “formidable”), not a single fixed idea.
• Founder quality > idea specificity:
  • Ideas may be “mutable,” but founder capability and ambition drive outcomes.

---

Key metrics / targets mentioned

• No explicit financial KPIs (e.g., revenue, CAC, LTV, churn) were provided.
• KPI named directly: shipping pace (iteration speed).
• Cost signal (qualitative):
  • compute/token burn, including very high daily token costs.
• Valuation reference:
  • Founders may become “billionaires” unexpectedly based on last-round valuation.

---

Actionable recommendations / playbooks (as stated or implied)

• Founder profile
  • Prioritize “formidable” traits: alignment via equity and ability to get what you want under adversity.
• Daily execution motivation
  • Treat fear of failure/disaster as a driver for focus and rapid response.
• Milestone-based funding with limited capital
  • Start small, build convincing artifacts (designs/simulations/expert reviews), then raise to the next stage.
• AI cost management
  • Expect major spend shifts from salaries to compute/token usage; build iteration around controlled burn.
• GTM via batch ecosystem
  • Use the batch as a built-in early-adopter base (“YC GDP”) to validate quickly.
• Operational cadence
  • Optimize for shipping speed—AI tools don’t eliminate the need to ship.

---

Presenters / sources

• Presenter: Paul Graham (Founder of Y Combinator / YC)
• Interviewer / host: Unspecified (a moderator/guest speaking from YC offices)
• Referenced sources (people/products):
  • Patrick Collison (commentary on “lean startups may be dead”)
  • OpenAI (used as an example in AI/Google replacement discussion)
  • Sam Altman (formidable-founder reference)
  • Sid (GitLab) and his essay: “I went founder mode on my cancer” (referenced)
  • Philip from StarCloud (rocket startup / white paper + booked launch example)
  • The Lean Startup (book referenced; PG says he hasn’t read it)
  • Turing Test (referenced for AGI discussion)