AppLovin CEO: Why Founders Shouldn't Angel Invest & Why the Best Don't Need Mentorship

Summary

• AppLovin’s financial profile has no comparable, and Foroughi knows it invites suspicion: ~$150B market cap, 84% EBITDA margins, rule of 40 running near 150, ~70% year-over-year growth, and reaching or over $10M EBITDA per head across the ~400-person core business. “There’s not another comp in the world that looks like it” — and his explanation for the short attacks follows directly: “in a world where things don’t make sense, people think you’re cheating.”
• The turnaround was a central conviction bet at the bottom: after falling 92% in 2022 to under $4B (under 4x EBITDA, while growing ~40%), he declared the old recommendation-system ML dead, slowed basically all R&D on it, turned over the people committed to it, and rebuilt on cutting-edge techniques — Model Axon 2, launched April 2023 — with the stock later running from $9 to $750 in roughly two and a half years.
• The 2022 buyback created “roughly a third of the company’s value… call it 50 billion around”: he shut down investor relations, raised some debt, and bought back specifically from the flimsy COVID-era cap table that needed to sell, removing the overhang. But he’s explicit that buybacks don’t usually pan out — “you sort of trade where you deserve to trade” — citing Wix’s big buyback followed by a ~25% weekly drop.
• The org call: he cut 40-50% of staff in most departments during a near-triple-digit growth year, rebuilding “as if we were building it knowing what technologies were available to us today” — HR went from 70-80 people to ~15. Same logic applied to AI spend: “token quotas and token budgets are no different than hiring quotas,” and he expects “a lot more tech layoffs over the next couple of years.”
• The SaaS apocalypse is fair and “I’m not sure it’s actually done yet” — LLM shipping speed makes terminal value “dicier,” and the SBC death spiral (3% dilution becomes 10% after a 66% fall) compounds it. Judge everything on cash flow minus SBC; AppLovin holds its grant flat at ~$300M a year. And on startups: “I would be very very nervous if I was building a business as an interface on top of” the frontier labs.
• Management heterodoxy throughout: no product org (engineers are the product managers; 80-90% of code is AI-written but “that discounts quality over quantity”), no one-on-ones, no reviews, little mentorship — “really good people figure out a way” — a four-person exec team, and Claude Code as the shop standard with Cursor “less so these days.”
• The personal ledger, stated without varnish: founders shouldn’t angel invest (the distraction losses “can compound”), kindness has a speed cost (“if you’re too kind and not as direct, not as aggressive, you’re wasting time”), and the price of the grind was presence — his kids’ childhood was “sort of a blur”: “I was there, but I wasn’t there mentally.”

Deep dive

1. Numbers no comp matches — “people think you’re cheating”

• The profile: 84% EBITDA margins, rule of 40 “in the last quarter, I think was like 150,” ~70% year-over-year growth, and a business “expanding without adding heads.” Of 895 total employees, the core advertising products run on ~400 people (likely Adjust analytics and likely Wurl, a CTV business, run separately) producing nearly all EBITDA: “reaching or over $10 million a head now.”
• His framing of why that draws fire: “A lot of the things that we’ve been able to accomplish just don’t make sense to people. And in a world where things don’t make sense, people think you’re cheating” — instead of realizing “you built one of the cooler technologies the world’s ever seen.”
• The short attacks (including a single -23% day) were almost structural: the stock ran $9 to $750 in about two and a half years, under $4B to ~$250B market cap — “I don’t know if any other company has ever seen that kind of value creation in that short amount of time in history” — with a “goofy name” and no story proactively told. “We were sitting ducks.”
• His gripe with the mechanics: short sellers can take large bets on puts, publish “overly dramatic” reports admitting “we’ve most likely covered our short position by the time you’re reading this,” and face no downside — while he’s SEC-bound to accuracy. The counter he owes his clients: well over $10B a year is spent on the platform on a performance basis, so attacking AppLovin “is effectively calling all these advertisers… a bunch of morons.”

2. Down 92% in a year — then throw out the whole stack

• The arithmetic of the trough: IPO’d in 2021 at $28B with $700M EBITDA, ran to ~$40B, then in 2022 cleared $1B EBITDA growing “40%ish” — and still fell 92% to a little under $4B, under 4x EBITDA. “To fall 92% in a year you go down almost every single day of the year… you got to go up 10x to get back to where you started. It is a bloodbath.” Stebbings’ aside from Neil M (likely Neil Mehta): the difference between down 98% and 99%? “Half.”
• He rejects the don’t-watch-the-ticker pose: “I 100% look at the stock price” — it’s real-time world opinion, and the real danger is contagion of doubt: “Am I doing something wrong?… is everyone else in the world smarter than me?” Employees’ families were asking why investors weren’t buying — “it’s easy to get tricked into believing it” when it falls that far.
• The bet at the bottom: the old recommendation-system ML was declared outdated, basically all R&D on it slowed, people committed to the old system turned over, and the whole thing rebuilt to the cutting edge — Model Axon 2, April 2023, with near-triple-digit growth in 2024. Executing it required voicing conviction while “people are calling you thinking you’re suicidal… you got to drown out that noise.”

3. The buyback that created ~$50B — and why buybacks don’t usually pan out

• Root cause of the collapse, in his telling: a COVID-window IPO that never built a blue-chip roster — “our cap table was basically the private market cap table that needed to sell.”
• The move: shut down investor relations entirely (“why do I need to go to a conference to explain to everyone who’s selling my shares to buy my shares?”), deploy every dollar generated plus some raised debt, and buy specifically from the sellers — company folks, ex-co-founders, investors needing liquidity — rather than shaving float pro-rata. Result, marked to today: “roughly a third of the company’s value came from that buyback… call it 50 billion around.” “It was a good buyback.”
• The warning attached — Stebbings’ phrase “it’s like a bridge round” left standing: “It’s easy when you’re inside to think you’re cheap, but you sort of trade where you deserve to trade.” Wix did “a big ass buyback” and dropped ~25% in a week; “none of us are day traders when we’re running businesses.”
• The IPO meta-lesson: going public is just “a fundraising route” like a Series A — “there’s no right time to go public… you can’t time the market. Just go public.” The standing obligation afterward: beat “the basket of the S&P” over every rolling three-to-five-year window.

4. The $83M package was a recovery option — and 2022 rebuilt the man too

• Stebbings’ number: $83M total comp in 2023, eighth-highest-paid CEO in America. Context: Foroughi had taken essentially no compensation since founding; at the $9 bottom he asked for a package for the first time, “to align myself with investors” — first threshold around $38-40, five or six levels up to $80, the IPO price (as spoken, “APO”), on a fixed term. “In order for me to get paid anything, the stock had to clear that and then keep going up from there.”
• His defense of founder comp: the never-pay-them-again logic is “completely flawed” — he was wealthy on paper and could have left to start a fourth company; big pay for big upside keeps builders committed. And the role is misread from outside: “It’s a brutal job. It’s lonely. It’s stressful.” Stock up, people “believe you’re smarter than you are”; stock down, they think “you might jump off a building.”
• The personal reset at the bottom: health decaying — eight coffees a day, losing hair, sleep, fitness — and a harder admission: “almost in every relationship in my life, I was never really present.” The fixes were small and concrete: 10-minute blocks of total presence with his kids, and learning to surf (“you have to put the phone down”). “By getting that back, I became a better CEO.”
• Delegation was the other 2022 commitment: a controlling CEO for a decade, until CTO Giovanni “took the product role away from me.” His read on Paul Graham’s founder mode: “an extreme reaction to extreme bloat” — once the team is lean and exceptional, delegation is powerful. He even handed the board chairmanship to Craig Billings, CEO of Wynn (likely): “if there’s someone better to do it, step aside.”

5. Layoffs into triple-digit growth

• Mostly in 2024, growing “near triple digits,” he cut staff 40-50% in most departments. The heuristic: ask what the company would look like “as if we were building it knowing what technologies were available to us today” — then jump straight to that org rather than let it arrive slowly.
• The sequence: eliminate processes he didn’t like, remove their gatekeepers, then hit automation-exposed areas — HR from 70-80 people to maybe 15 (keeping the individual-contributor doers), creative production (AI produces, fewer humans innovate), and engineers who only get “a 2x instead of a 10x or 100x” from the tools. Keeping people in dead-end roles, he argues, poisons A-player morale — and “A players won’t exist in bulk if you have a bunch of B’s, C’s, and D’s around them.”
• The employee deal is stated to their faces: “You become AI native, you’re going to have a role here. If you avoid utilizing these technologies… you’re going to get fired and that’s life.”
• On today’s layoff wave: it’s COVID overhiring, not AI — “the former is still yet to take full effect” — and cuts may not deliver, because in a bloated company “your A’s are probably already long gone”; fire half of mediocrity and “you’re left with half mediocrity.” The only real fix: “fire 99% of people and just rebuild it from the ground up” — nearly impossible at a public company. His call: “a lot more tech layoffs over the next couple of years.”

6. Doers only: no CHRO, no one-on-ones, little mentorship

• The exec team is CEO, CTO, CFO, general counsel — no COO, CRO, CMO, CHRO. Stebbings, deadpan: “I’m shorting the company.” The catalyst was Giovanni arriving and asking of everything, “why do we have these people, why do we have these processes” — which sent Foroughi back through the whole org to the founding culture.
• No one-on-ones, no reviews: criticism arrives real-time via chat, and approval is implied. “Really good people figure out a way. They don’t need a whole lot of mentorship… good people don’t need that type of handholding.” People who need heavy development aren’t the people he wants.
• No formal learning and development either — instead, document everything in Slack and transcribed calls so any new hire can ask: “Hey, Claude, summarize for me what Adam cares about over the last quarter and write me a book of everything that matters to him.” In-person is used for key clients — “you can’t replace in person,” and “as we go to this world where bots are going to do more for us, in person’s even more valuable.”
• Stebbings’ pushback on socializing — “can we not bond over a whiteboard… why do we have to go and sit and drink in a pub?” — gets a real answer: with your best people you get “heated debates, like yelling matches,” and without dinners and drinks “resentment can build.” Also: “sometimes your best ideas come out of those moments” of getting drunk together talking about work.

7. 80-90% AI-written code — and token budgets are the new hiring quotas

• Likely Databricks announced 50% of its code is AI-generated; AppLovin’s is “80, 90% probably — but that discounts quality over quantity.” Chasing token metrics produces slop plus “massive fees to go pay the large language model businesses.” His test: “was your investment in tokens covered by the amount of revenue that you created from the code contributed.”
• Why AppLovin can measure that: no product organization — engineers are the product managers, and a model improvement shows up with certainty in accuracy numbers and revenue. His forecast for the product function everywhere: “either your product people become engineers or your engineers become product people, but you don’t need both” — though engineers are still needed to audit code for security and slop.
• Tooling datapoint for the picks-and-shovels watchers: “most people are on likely Claude Code,” likely Codex is used as well, “and Cursor less so these days.”
• The line to keep: “Token quotas and token budgets are no different than hiring quotas. Until they get efficient, they’ll be inefficient and I think a lot of companies will just burn money” — the same dynamic as Valley hiring quotas over the last 15 years.

8. Don’t be an interface on frontier models

• Asked whether most companies founded today get commoditized by the labs: “I would be very very nervous if I was building a business as an interface on top of those companies” — you’d “better build a moat really, really fast given how exceptionally talented companies like Anthropic are about releasing product on top of their own models.”
• AppLovin’s own approach is a purpose-built recommendation system — the model family behind Facebook’s, TikTok’s, and its own ad systems. You can’t just ask an LLM “what’s the next ad to see”; “that wouldn’t work as well as a custom model built for this purpose.” On TikTok’s engine, he sidesteps the targeting claim and praises its engagement: “quite phenomenal.”
• Contra likely Elad Gil’s tweet that compute is the currency for talent: it depends on the space. Anthropic is “doing the best in terms of releasing models and product” yet “probably does not invest the most in compute” — culture, people, and focus did it. Recsys researchers are “not bound by compute, they’re bound by curiosity.”
• On security (Stebbings cites the Lovable and likely Vercel breaches and asks whether “models like the likely ‘Mythos’” will unravel previously unseen vulnerabilities): the risk is real and shipping speed makes it worse first — “you’re going to have more security breaches most likely” — while he hopes model-driven code auditing eventually produces a more buttoned-up equilibrium.

9. Judge everything on cash flow minus SBC — SaaS pain isn’t done

• AppLovin grants a flat ~$300M of stock a year against a ~$150B market cap. After 2022 it restricted grants to the top 10-15% of the company; everyone else takes cash plus an ESPP option — because someone living paycheck to paycheck can’t absorb a 92% drawdown in their comp. “Companies tend to give away their stock too cheaply and too broadly.”
• Why the metric: “cash is king.” A company generating $1B of cash while issuing $1B of equity and buying it back “is not generating any cash… what’s the real value of that business?”
• The SBC death spiral he sees in fallen software: a 3%-of-cap-table burn becomes 10% after a 66% decline — “a level of dilution that it’s incredibly hard to come out from underneath” — which itself makes the stock hard to bet on.
• So the SaaS apocalypse is fair — “and I’m not sure it’s actually done yet.” The rapid rate of LLM product delivery makes traditional enterprise SaaS “hard to bet on years into the future,” so terminal value gets “dicier” and investors sell. He doesn’t expect a wipeout — “companies once they’re embedded with you utilizing a certain software usually don’t change” — but “a lot of the growth opportunities are gone.”

10. The trillion path, and the trades he’s made in life

• Path to $1T without a social network: “if we ever got to generating $30, $35 billion of cash a year, we’d probably be a trillion-dollar business.” Levers: better monetizing the billion-plus daily-active-user gaming audience (“adult audience, a lot of heads of household”), and CTV — “one of the holy grails of advertising” — porting mobile performance ads to television for SMBs. An engagement model is interesting as “a talent-recruiting play… not a requirement.”
• Winning over money, with the receipts: he didn’t cash out after a 2015 all-cash offer “in the hundreds of millions” because earlier exits meant “my bank account was sound… this had to be the home run.” Yet the doubt never leaves: “that fear of blowup is one of my big motivators” — he still wakes up checking stats, wondering “are we going to go bankrupt today?”
• Why he says founders shouldn’t angel invest: you have to sell your own shares for liquidity, then chase a second KPI — and the distraction loss is unmeasurable: “as those losses start adding up, they can compound.” “Every second of my available time should be committed to” the company.
• On kindness — the episode-title claim, unhedged: “if you’re too kind and not as direct, not as aggressive, you’re wasting time.” His own reference checks: half say aggressive and like it, “half the people will say I’m an [asshole]. They’ll all say I’m competent.” And the cost, stated plainly: his kids’ growing up was “sort of a blur” — “I was there, but I wasn’t there mentally. It is not a great thought when you have that.”