Hard truths about building in the AI era | Keith Rabois (Khosla Ventures)
In a Nutshell
Keith Rabois argues that AI obsoletes traditional PM roles, shifting leadership to CEO-like skills in deciding what to build and why, while top organizations prioritize "barrels" (independent initiative-drivers) over headcount, ruthless talent density via internal promotion and undiscovered gems, and relentless velocity without psychological safety or customer interviews (except enterprise). Success demands iPad-only workflows, public criticism for high performance, and accumulating advantages like network effects in AI startups. Key tactics: 20+ references per hire, 30-day post-hire feedback, and spotting undervalued talent overlooked by big tech's processes.
These notes were generated by AI and may contain inaccuracies.
Idea of a PM makes no sense in the future. The skill is more like being a CEO now: what are we building and why?
There's a lot of anxiety in the job market. AI is going to radically reorient lots of people's careers.
In the best organizations, the number one consumer of tokens is the CMO. They don't need to rely on deputies to get actual work product.
Contrarian takes: don't actually want to be talking to customers. "I hate talking to customers. I refuse to allow colleagues of mine to talk to customers."
Criticizing in public versus in private. High performance machines don't have psychological safety. They're about winning.
Keith Rabois: early investor in Stripe, Palantir, Airbnb, YouTube, DoorDash, Ramp, and dozens of others. Part of the PayPal mafia where he was executive vice president of business development and policy. COO at Square, VP of corporate development at LinkedIn. Co-founded two companies. Currently managing director at Khosla Ventures.
In the 99.9th percentile at identifying talent, building teams, and operating world-class companies.
Keith does everything from an iPad, phone, or watch. Hasn't touched a computer since September 2010, when he started at Square and saw Jack Dorsey running the company off an iPad.
Benefits: avoid distractions, flexibility, screen angles, light weight. Carries iPad everywhere. No reason to use a heavier, less flexible machine unless doing heavy-duty engineering.
Engineers starting to code from their phone talking to AI. Keith has been preparing for this by following Jack Dorsey, who is ahead of the curve.
"iPad maxing. Keith Raboy. See if you got my Apple. As long as it's an Apple product, it works."
Lots of people struggling to find jobs amid layoffs, but harder than ever to attract best talent with amazing companies, money, $100M offers.
From companies best at attracting top talent:
Start with first principles: "The team you build is the company you build." Learned from Vinod Khosla at Square. If right people, everything else easy; wrong people, everything difficult.
PayPal success due to Peter Thiel and Max Levchin marshaling incredible density of talent. Led to generations of successful companies (25 years).
Focused on critical density of talent: identify, retain, promote.
Early career at PayPal: not great at hiring strangers (50/50 hit rate). Learned by stealing talent internally from underleveraged people. Got feedback from David Sacks (COO): need leadership leverage where 1+1=3+.
Lesson: could identify talent with context (lunch, dinner, runs). Double down on people known. Later developed techniques for strangers.
Advantage for founders: if shows ability early to assess talent ruthlessly and accurately, can go very far with no other abilities.
Hiring is a muscle: exercise it, learn from feedback.
Tactics improve you in the middle of the bell curve (10-20 degrees). For top 10-50 basis points, deviate from playbook.
Teachable tactics:
- Ruthless referencing: Tony Xu at DoorDash does 20 references per senior hire.
- David Sze (Greylock, board at LinkedIn): reference until hit a negative.
Example mistake: Fair (co-founded by Max R and Jeff Cverson, ex-Square). Wrong question: "Was Max a good employee?" (mixed). Right: "Is Max capable of being a world-class entrepreneur?" (yes). Many VCs passed.
Helpful questions:
- For candidates (senior/leadership): "If you were CEO [of their current company], what would you have done differently?" Reveals strategic mindset, value creation.
- Follow-up: "Why weren't you able to persuade [CEO] to do it?"
- For references: "What would lead to this person being most successful? If not working out, primary root cause?"
Feedback loop: 30 days after hire, ask entire team: "Would you make the same decision?" As accurate as 1-2 years out.
Most companies raise seed, get traction, raise A/B, hire a lot → burn rate up, but less/more work done. CEOs frustrated.
Fundamental driver: limited barrels (people who can independently drive initiative from inception to success).
Hiring more without more barrels = stacking people behind same initiatives → collaboration tax, coordination tax, drag coefficient.
Examples:
- PayPal (254 people in Mountain View when acquired): 12-17 barrels.
- Glatir (Jack Altman): 2 barrels (common for good company).
Barrels dictate parallel vs. sequential initiatives. To do more: need more barrels.
Ratio of barrels to ammunition dictates simultaneous important initiatives.
Need ammunition too (designers, engineers, PMs, data analysts) depending on project. Be deliberate about ratios.
How to identify barrel: Can take idea and make it happen. "There's a hill over there. Get us over that hill." Motivates people, accumulates resources, measures. Delivers outcome come hell or high water.
Skills: internal pol org, resource strategy → collection for outcome.
Agency: takes off plate, fire-and-forget. Proactively returns with issues tried, diagnosis, asks for help with time to intervene.
Example barrel: Intern Taylor Francis at Square. Keith wanted 9pm cold smoothies for late-working engineers (healthy, not junk). Office team failed. Taylor (day 2): "I'll solve it." Delivered on time, cold, great on standing desk. Keith gave him everything to do.
Standard: mission, selling the vision (indispensable).
Nuanced: Convince their skill overlaps with critical blockers. They're betting on themselves. E.g., if superb at marketing and that's the blocker (not tech/product).
It's easy to recruit a world-class marketing person by emphasizing that the company is building something cool they'll be proud of, and their unique ability can solve a specific problem. Keith Rabois joined Square in 2010 this way. After being acquired into Google, he planned to become a VC post-vesting. Square investors contacted him, saying they'd searched nearly a year for someone who knows financial services yet is entrepreneurial—claiming only two or three such people existed worldwide. This convinced him to leave Google after two weeks, delaying VC plans by three years, due to the ego and impact of being one of the few who could do the job.
You have talents. You want to use them and feel challenged every day and that what you're doing matters.
Build companies on undiscovered talent rather than competing for everyone else's top picks. Peter Thiel taught Rabois this first week at PayPal while jogging Stanford campus: find undiscovered talent to scale against incumbents with infinite money. Rabois has followed this for 25 years. See his RAMP speech on hiring and Eric Glyman's (Ramp CEO) similar speech, both online.
Startups can't afford top-company talent who did the exact thing, and they're often not ideal for early stages. Like sports with salary caps, startups have one-tenth the cap of competitors, so leverage fewer assets for more success.
Isolate why large organizations' homogeneous recruiting processes miss them. Predict what Meta, Google, Block, or Coinbase would overlook. Often due to lack of data: younger people have fewer data points, making black-box evaluation hard (like FICO scores). Over 30, abundant data leads to standardized processing; no data creates alpha.
This mirrors investing skill in picking undervalued startups.
Rabois is a bar raiser, pushing harder no matter the numbers—especially when succeeding. A CEO friend asked Mike Morita the common denominator of best CEOs: relentless application of force. CEO's job offsets complacency, which grows with success. Most businesses lack network effects and can't autopilot.
When struggling, be supportive/coach-like since founders know issues—criticism doesn't help. When thriving, be critical to spot future problems amid happiness/complacency. Like sports: polish details when winning; excite about future when losing.
Advice for founders/product leaders: keep pushing harder, raise the bar even when doing great. Success creates inflection (valuation, momentum), but like winning Super Bowl, reset to 0-0 next year. Companies can skate briefly; VCs can't—Rabois must daily find undiscovered founders despite 13 strong years.
Brian Chesky (Airbnb) exemplified pedal-to-metal always; brief breaks dropped morale as top talent craves creation/value/drive. Talented people are like superb athletes—coasting lowers their morale.
AI will radically reorient careers. Thrive via intellectual curiosity: learn new things to embrace future. Beyond working harder (Tyler Cowen: work harder as AI erodes value), curiosity future-proofs. Business people (like Rabois) can now use AI directly.
Top token consumers: CMOs at Opendoor and another board company—despite many engineers. They build analytics, campaigns, landing pages, A/B tests without deputies, shipping drafts/insights directly to CEO.
Peter Thiel convinced Rabois PM role makes no sense: PMs take customer inputs for year-long roadmaps—ridiculous now with foundation models (e.g., Lovable) advancing rapidly. Impossible in November, easy by March. Need organizations adapting roadmaps on-the-fly.
Future: notice/exploit weekly possibilities for next-week features/value. Conventional PM intermediaries obsolete; need intellectual preparedness.
AI eats software middle, enabling anyone to build via prompts. Gap: figuring out what to build and aligning around it. Skill like being a CEO: what/why. Tools ease creation; art is business acumen—understanding company equation, inputs/outputs to move needle.
PMs called mini-CEOs; great PMs/engineers/designers will thrive, but exceptional ones with commercial instincts premium. Examples: Max Levchin, Jeremy Stoppelman (PayPal engineer to Yelp founder, promoted to director/VP). Engineer + business-minded = ultimate unicorn.
AI premium on engineers shipping personally at scale: Ramp engineering director ships as much code as solo while managing 20, using AI as "second team".
Design/code merging—unclear if code becomes design or vice versa. Bets on both.
Design roles flat past three years. No time for traditional process amid engineers shipping 17x/day. Shopify (2+ years): no PowerPoint/Keynote product presentations—only workable demos. PMs build; execs reject static mocks.
Design alpha like marketing: storytelling to cut through clutter (NP-hard: best framing amid infinite options). As AI expands build/distribution, core remains what to build + getting attention.
Customer feedback flawed—not real-world clutter (e.g., while at Barry's Bootcamp, working, parenting). Isolated experiments mislead.
Similey simulates humans (AI models of people) for pre-launch marketing/product tests. Question: training data quality?
AI content will surpass human—inevitable. Chinese TikTok: all AI videos, highly enjoyable. Binary sort: human-generated thrives with provenance/authenticity premium (e.g., Warhol art valued despite AI replication). Curated human experience as premium.
Content will be ranked by quality regardless of whether it is human-created or AI-generated. This mirrors Ben Thompson's trajectory of curation versus algorithm: curation for human-created content, algorithm for the best overall.
AI excels at generating fun-to-watch videos and high-quality images, but writing remains poor despite being called large language models trained on text. This is ironic as text is their weakest area.
LLMs ration tokens economically, producing better quality for short prompts than long ones (e.g., short piece vs. page, chapter, or book). Quality degrades with length due to token rationing versus actual capability.
Effective writing process: short examples generated by AI, then humans pick, edit, and reprompt. Echoes Hemingway: shorter writing takes more work.
As a litigator 25+ years ago, the hardest part of a brief was the first paragraph. Nailing the first three sentences dramatically increased chances of winning. With three weeks for a brief, first week spent framing those sentences—often epiphany in shower or run—then 30 pages in two days.
AI handles the 'power through' part well but still needs the strong opening.
First 4.5 years as law clerk and litigator, useful for trading business vs. legal risk. Investments in financial services/heavily regulated areas (e.g., YouTube) benefit from legal risk assessment.
Lawyer training inconsistent with entrepreneurship:
- Graded on issue spotting: identify all issues and resolve—focuses on what can go wrong, not solving as entrepreneur.
- Productivity measured by hours worked/billable hours. Took 1.5-2 years post-transition to tech to stop time-tracking everything.
David Sax and Peter were lawyers; Peter quit after 5 months and 4 days; David never practiced.
Fewer contrarian takes now because they've been proven right. Good ideas should be adopted, like investor wanting contrarian start (Airbnb) becoming consensus for success. Need refresh rate on ideas.
Advice: Unless building enterprise, don't talk to customers. Refuses to let colleagues do it. Customers don't know what they want; subconscious decisions (e.g., purchases, clothing) lead to misleading conscious answers.
Proverbial example: Ask Porsche/Lamborghini owners why they bought—99% cite every reason except the real one (status?).
Enterprise works: decision-maker makes utilitarian decisions; extract info valuable despite politics.
Consumer/SMB/micro-merchant: unmitigated disaster.
Rely on instincts, gut, experience. Everything important about humans in Shakespeare—better than customer research.
Movie analogy: Optimize for selling tickets (PMF). If failing, check trailer, distribution; iterate on casting/script unlike movies.
Customer talk harmful: Locks in non-representative feedback (e.g., 'talked to 8 customers, not statistical, but...'). Can't unhear it.
Enterprise exception: 30 must-win accounts—talk to all decision-makers/CEOs.
For billion-person target: not representative.
Failed companies result from it. DoorDash: Customers didn't request food delivery button. Tested restaurants on placards, density for break-even deliveries.
93% US restaurants don't deliver—should be higher than 7%. Andrew Mason: Phones need 'I'm bored' and 'I'm hungry' buttons. DoorDash = 'I'm hungry' button. Insight made economic scalability key.
Applies to SMB-ish like DoorDash, Square—sub-mid market directionally dangerous.
Best companies have foundational insight; acid-test/pressure-test logic.
Airbnb: Brian Chesky showed Craigslist listings for 'rent my bedroom'—30 in Bay Area despite typing effort = real market. Leaned in after that.
Solving own problem key; sampling 10 random people risks 9+ saying no.
Taylor Swift's speech: Started alone iterating songs/piano thousands of hours vs. today's constant posting/feedback. Advice: Avoid exposure early.
Artists struggle recreating success: First was instinct-driven; later feedback creates derivative works.
Chain Smokers podcast with Jack Altman/Alex: Song resonated with new audience, not core—trapped by success.
Ugly baby from Creativity Inc. (Ed Catmull/Pixar): Great ideas start unwanted.
Startups same: For seed/Series A, want half VC friends laughing—ugly baby = real alpha.
Early employees at generational companies (OpenAI pre-fame, Palantir, Stripe): Sought ideas people laughed at as crazy.
Keith's parents laughed at most jobs (e.g., at tech); appreciated Stripe, lobbied to invest.
Existential question: Foundation labs so proficient no oxygen left? Startups need durability 8-20 years.
Key question: Accumulating advantages over time (e.g., network effects making business better/easier).
At seed: Founder articulates potential conceptually/sequentially (don't need empirical proof; rare every 5 years).
Founder-driven investor: Non-zero chance of changing industry/world? For seed/Series A, invest. Others: tech-driven (Marc Andreessen), product-market (David Weiden, Alfred Lin), mixes (Navis).
Extraordinary founder = next Brian Chesky.
Subtle early signal: Speed/operating tempo. Identify opportunity/problem at board meeting, ship solution by next.
Square: Roelof Botha (Series B lead) after 6 months/two meetings: 'Haven't seen this tempo since PayPal.' After seed-Series A tight window, Delian Asparouhov (chief of staff): Pace would make him leave VC.
Fair: Identify issue morning meeting, ship/measure by next.
Ramp: Led seed May 2019, preempted Series A September. Precipice of shipping cards in 3 months (vs. usual 9-12): program managers, sponsoring bank, etc.—unprecedented velocity.
Density of talent: Team improves (e.g., DMs get deeper).
Hiring philosophy: Skip senior experienced people; promote internally-grown talent. Works for Ramp, Trade Republic.
Exceptions exist; not hire GC right out of law school (did once, worked).
Delian's blog (lessons from Keith): Hiring senior for value preservation (sub experience useful) vs. value creation (not).
Step two was talent density and three is helping people develop into the role versus finding someone. Speed trickles down from who you're hiring. People use chief of staff roles to groom talent. At one phenomenal company, the current CMO was the previous chief of staff and the new head of product is the current chief of staff. This creates a factory absorbing ambitious talented people, training them over one or two years through osmosis to become senior successful leaders.
When talking about speed, consider ramp. Ramp's CPO emphasizes velocity. They have days.ramp.com tracking days since launch. Every board meeting starts with the first slide: day 1184. They're worth a reasonable fraction of a hundred billion.
One hot take: criticize people in public. Learned from a great founder. Logic: giving negative feedback individually optimizes for the atomic unit, not the system. Public criticism ensures all colleagues know the issue is being addressed, dispelling suspicions or concerns. It allows others to offer help, turning it into a team-building exercise rather than telling one person to fix it alone while the company remains nervous about persisting problems.
It may feel aggressive. Best sports coaches mix public and private feedback. Focus on winning. High-performance machines don't prioritize psychological safety; they're about winning.
If you want to be Michael Jordan, you got to act like Michael Jordan.
Read Jordan Rules or watch The Last Dance.
Psychological safety is negatively correlated with success for the most part.
Many failures along the way despite successes like PayPal.
Aqua-hired into Google and stuck there—not successful. Sold a company for 187 million but nowhere near the ambition; didn't achieve product or company goals.
Investing: world-class early-stage investor has 30-40% hit rate (golden); 50-70% failing. Like Michael Jordan missing 109 game-winning shots or Federer winning 60% of points but losing 40%.
Art of not getting too caught up in failure. Recently gave feedback in a board meeting: well-meaning board members wanted retros on failures, but the company is doing well. Avoid deterring ambitious shots on goal. Overemphasizing failures makes people fear criticism. Encourage risk-taking and excitement for difficult problems to create value. No retros; focus on winning.
Books
The Upside of Stress by Kelly McGonigal (Stanford professor). Argues you need more stress for happiness, health, or wealth. Compelling evidence at outcome and biochemical levels; transformative.
Recent Movie/TV
Nuremberg trial (TV). Lessons applicable to modern world. Learned 5-10 new things despite studying history and politics. Extremely well-produced; useful for understanding history's travesties and prevention. On Netflix or iTunes or both.
Product
Eight Sleep—prioritize 8 hours sleep even when busy. Investor, so biased, but transforms lives; still addicted.
Life Motto
No days off. From Bill Belichick starting Patriots' back-to-back Super Bowl championship parade chant. No days off for workouts or work. Missed only 7 workout days in 7 years (still annoyed by half; reoriented schedule for most). Last year: zero misses. Proxy for no excuses. Works out every day, typically more than once (e.g., Barry's class morning and later, plus lift). Includes illness, travel, weddings, international time zones.
PayPal Mafia Overperformer
No one overperformed unexpectedly. Invested in most derivative companies; had good sense for founder-level ambition.
Find Keith at x.com; tweets prolifically (check pinned tweet).
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
