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

Key points

• **The 2022 reset playbook:** at 92%-down, threw out the entire ML stack, fired team members tied to the old system, rebuilt 'Axon 2' (modern recommendation system, post-LLM-research-era). Stock recovered ~$9 → ~$750 (≈80x). The hard call wasn't the layoffs — it was halting *all* R&D on the working old system to commit to a new architecture.
• **EBITDA-per-employee ~$10M, 84% EBITDA margins** in the 400-person core advertising unit. Foroughi: 'there's not another comp in the world that looks like it.' Direct evidence that AI-native operating leverage produces financial profiles that traditional analysts call 'cheating' (and short-sellers do call it cheating — repeatedly).
• **Cut 40-50% of headcount in a near-triple-digit growth year.** Logic: 'rebuild the org as if we were starting today knowing what technologies exist.' Eliminated CMO, COO, CRO, CHRO, Chief People Officer — has only CEO/CTO/CFO/GC. HR went 70-80 people → 15. Forcing-function rationale: A-players don't want to be slowed down by B/C-players who fight automation.
• **80-90% of code is AI-generated.** Caveats it heavily: 'percentage of tokens consumed' is a slop metric. Real metric is whether the token spend produced revenue. Stack: **mostly Claude Code, Codex used, Cursor less so these days.** This is a high-signal data point on what a top-decile lean operator actually picks.
• **No product managers.** Engineers are PMs. AI-native engineers must own KPI ownership directly. Forecast: 'product as a function disappears — your product people become engineers, or your engineers become product people, but you don't need both.' Direct contradiction of bucket-2 system-of-record SaaS that depends on PM/engineer separation.
• **Cash flow minus SBC is the only valuation metric.** Companies are dilution-spiralling: stock falls 66% → SBC was 3% of cap, now 10%, can't escape the dilution. AppLovin's SBC ~$300M/yr on $150B market cap = trivial burn. Direct call-out of Wix, ServiceNow, Salesforce, Databricks-style SBC heavy comp.
• **The SaaS apocalypse 'is not done.'** Foroughi: 'when you get into an unpredictable outcome in the future, it's very easy to sell businesses. The rapid rate of product delivery in the LLM space makes traditional enterprise SaaS hard to bet on years into the future.' Companies don't wipe out — embedded software is sticky — but growth opportunities die, terminal value gets discounted, SBC % blows out, downward spiral. **Operator-side validation of Lemkin's three-bucket framework from 20VC episode.**
• **Buyback masterclass (and a counter to the buyback wave):** at the bottom, didn't buy from open market. Identified Covid-era forced sellers (private cap-table that needed to exit) and bought *directly* from them. Took out the structural overhang. Estimates this single move contributes ~33% of today's $150B market cap (~$50B). General buyback advice: don't do it unless you can identify a structural seller block — Wix-style 'big buyback then -25% in a week' is the failure mode.
• **Token-budget anti-pattern.** Companies are putting people on token leaderboards = same mistake as hiring quotas. Engineers ship slop to top the leaderboard, real revenue doesn't move. Right approach: tie token spend to a measurable revenue KPI, then *invest* (don't budget) in tokens.
• **Compute-as-talent-currency is space-dependent** (responding to Eli Gill tweet). LLM labs need compute to attract researchers. Recommendation systems don't. Notes Anthropic 'probably doesn't have the most compute' yet leads on product — culture and people did the work.
• **Founders shouldn't angel invest.** Two reasons: (1) requires selling shares of the core business; (2) divides attention. 'My goal is to make the company as good as it can possibly be 10/20 years from now. Every second of available time should be committed to it. I don't know what the loss is when I get distracted, but the losses compound.' Direct contradiction of the founder-as-VC trend.
• **Path to $1T market cap is not social network.** Need ~$30-35B/yr cash flow. Get there via: better gaming monetization + Connected TV performance ad port (the 'holy grail of advertising' = small-medium business performance ads on TV). Optional adjacencies (engagement model, social network) are talent-recruiting hooks, not core thesis.
• **On Mythos / cyber risk:** acknowledges Anthropic may be slowing rollout for compute reasons or for safety reasons — 'somewhere in between.' Short-term more breaches as faster shipping outpaces audits. Long-term: 'a lot more buttoned up.' Same upgrade-cycle thesis as Sacks on All-In.
• **'A players won't exist in bulk if you have a bunch of B/C/Ds around them.'** And: once a company is bloated, you cannot reverse without firing 99%. Most layoffs you're seeing today won't get the bloated incumbents back to high-output. Implication: incumbents that haven't reset are stuck at mediocrity-equilibrium.

Notable quotes

Themes

• Operator-side proof of the SaaS apocalypse three-bucket framework
• Lean AI-native ops produces financial profiles analysts call 'cheating'
• Layoffs in growth years — the radical AI-native reset
• Cash flow minus SBC as the only valuation metric that matters
• Token-budget anti-pattern and the slop economy