myainote

In Q3 2026, Coinbase transaction revenue fell 44% while stablecoin revenue fell just 5.4%, and Robinhood earned more from prediction markets than from crypto trading. A four-layer way to read where crypto companies' profits actually come from.

In December 2020, seven people left OpenAI to start Anthropic. What each of them does, how they wrote safety commitments into the company's rules, and what the IPO filing says about votes, board seats and their giving pledge.

One EIP-7702 authorization emptied a Binance hot wallet in 3 min 24 s. Binance can cover the loss; what broke is the industry's assumption about hot-wallet risk: max loss moved from one transfer's limit to the entire balance. ~$2.5B in stablecoins was frozen; ETH and 18,270 unmoved BTC were not.

Returning to the opening question — is SaaS still a compounding machine? — the honest answer: yes, but only for the few with high-quality retention, a deep moat, real cash flow, and pricing power. Its foundation is being shaken because AI hits all three pillars at once: near-zero marginal cost (inference has a cost), per-seat pricing (agents dissolve the "person"), and the entry-point moat. The truly scarce asset returns to the moat AI can't copy or bypass.

Tools aren't ability — this chapter integrates the book into a lifetime research system in three steps. One: screen with an eight-dimension scorecard (TAM, growth quality, retention, moat, profit/cash flow, management, AI, valuation) — NOT a "which to buy" formula; its value is in the filling-out and reading the weakness, not the total. Two: make research a process — a quarterly tracking sheet plus a written thesis, key assumptions, and kill criteria (the system's moral core).

The book's framework applied to nine real companies by thesis type: long-term compounders (ServiceNow, Veeva — biggest risk is the buy price), high-growth/high-valuation (CrowdStrike, Snowflake — where "good company, bad stock" concentrates), platforms (Microsoft, Salesforce — fate hinges on the will to self-disrupt), vertical + payments (Toast), turnarounds (Okta, Twilio — repair, not growth), and the value trap (a profile, not a verdict). Learn the method, not the verdicts.

A good company is not the same as a good stock — between them stands price, and the expectations baked into it. Your return depends not on whether the company does well but on whether it beats what the market already expected; so a great business priced for flawless execution (price ≥ bull-case value) becomes a bad stock. The usual trigger is slowing growth, which has two faces: temporary deceleration (law of large numbers/cycle — can create a buying opportunity) vs. permanent decline.

Valuation is a disciplined art, not an exact science — the goal is a framework, not false precision. Three rulers: EV/Revenue (most common, most dangerous — ignores revenue quality), EV/FCF and P/E (stricter, but use FCF net of SBC as a stress test), and reverse DCF (the book's favorite — asks what the current price already implies). The multiple a company deserves prices in everything from Parts One–Four, so it's the conclusion of the analysis, never the start.