Top 1% of startups spend 8.3x more on AI per employee than the average

3 trends in startups/tech/venture

📈 Top 1% of startups spend 8.3x more on AI per employee than the average

• Per the latest data from OpenAI, “frontier firms” (defined as those in top 10% of heavy AI usage each month) generated 8.3x as many output tokens per active user as “typical firms” (defined as those in the 50% of AI usage each month), up from 2.6x in January.

• Per Ramp, the top 1% of US businesses spent $7,400 per employee per month on Al last month, compared to $650 per employee per month for the top 10% of businesses and $11.95 per employee for the median business.

• This is in line with what I’m seeing from startups — the best performing startups tend to also be amongst the heaviest users of AI. Here is a builder index, where you can see how people are actually using AI — shoutout to DROdio, a longtime friend and portfolio founder, who often tops the list.

• The takeaway? Employees at the most cutting edge startups don’t just use more AI — they use a lot more AI.

🤖 Tech competes for access to proprietary enterprise data

• Spirit Airlines officially went out of business in May. While rival airlines quickly scooped up its airport slots, tech companies have been bidding on an asset they find much more valuable — corporate records dating back to 1986.

• Specifically, Spirit’s asset pool includes documents, workflows, spreadsheets, 100M emails, 500M Microsoft Teams messages, 7.5B anonymized transaction records and more.

• Tech companies prize bankrupt company data because authentic human business decisions provide far better training signals than synthetic or generic text. As a result, it was no surprise when Google won the bid for Spirit's 34 years of real-world workflows and operational archives for $10M, beating out rivals in their attempt to collect better data for training enterprise AI models.

💰 Secondary markets hit record highs

• As more startups remain private for longer, more employees and early investors are eager to cash out on their locked up equity. To address the pent-up demand for liquidity, companies have been increasingly facilitating “tender offers” — a window of opportunity for outsiders to buy a company’s stock off of existing shareholders at a set price.

• These tender offers have led to a boom in the secondary market. Per the latest data from Carta, tender offer activity has reached a four-year high, with the total value of equity changing hands tripling from last year to this year.

2 theses on what’s next

🤖 “Perfect” usage of AI = not too much and not too little

• Using too little AI could kill your startup, as you under-automate (and let your competitors execute faster than you).

• Using too much AI could also kill your startup, as you over-rely on AI (and become too generic to be useful).

• So, what does the “right” amount look like? Typically, it means delegating 80%+ of your existing workflows to AI, and then investing your newly-recovered time on human-to-human relationship building.

• This was one of the top topics discussed at the Jeffersonian dinner I hosted with Deb, Board Member of Poshmark and Intuit, former CEO of Ancestry, and longtime product leader at Meta. You can check out photos from the dinner, video of the public portion of the discussion, and consolidated notes from the entire dinner discussion.

🤔 Hire based on talent, not what’s open

• It’s not about roles. It’s about people. I believe leading startups and tech companies will increasingly move away from narrow job titles and descriptions, and instead focus on attracting/hiring versatile and high-caliber talent, regardless of open roles or titles.

• I think organizations will increasingly adopt the generalist/polymath "Member of Technical Staff" (MTS) model to deploy adaptable problem-solvers across multiple domains — prioritizing raw capability, agility, and cross-functional skill over predefined corporate functions.

1 tool I love

🎓 The U.S.’s first AI-enabled collegiate university

• Employers are spending more on AI training than ever before, but employees only complete self-paced AI courses ~12% of the time.

• So, what gets employees to finish versus abandon training courses?

• It turns out most employees:
  ❌ Will NOT bother finishing modules that lead to an internal badge, but ...
  ✅ WILL finish modules that get academic credits toward a real degree.

• Learn more about why degree-linked training outperforms stand-alone courses from Woolf.

More of my work 👇🏼

🎤 Decode videos | For top Berkeley and Stanford founders

• DECODE is the largest founder community co-hosted across UC Berkeley and Stanford. The DECODE annual conference focuses on helping founders in their earliest stages of starting a startup.

• I’ve had the honor of serving on the DECODE board since 2016. You can find videos from the latest conference I helped host here 👇🏼

🤖 Founders' prompt library

• My team and I have crowdsourced the largest AI prompt library optimized for founders – especially fractional founders — to help them build, sell and operate 10x faster and better.

👥 Pop up board

• I regularly host pop-up boards, a unique 45-minute session where founders ask their toughest strategic questions and get tailored advice from top builders and operators from Google, Microsoft, Meta and more.