.@garrytan’s Startup School talk is worth the time if you underwrite AI in public companies. He is not arguing that one person firms are on the way. He is describing the operating model that is already producing the numbers. Most valuation work still treats headcount as the scarce input. That is the wrong constraint. The scarce input is a process written down tightly enough that an agent can run it every day, and a simple router that assigns work to the right process. You do not need the whole warehouse. You need one shelf that compounds. One person plus agents is already an organization. You can run a real company before you incorporate, hire a cofounder, or write a deck. Tan’s examples are useful because they are specific. @mukundjha and @madhavjha at @emergentlabs hit $15 million annualized at 15 people, went from public launch to nine figures in eight months, and now sit above $120 million at a $1.5 billion valuation. @retellai reached about $60 million with about 40 people. That revenue per employee is outside the old software, oil, and railroad ranges. These are not outliers. In the current @ycombinator batch, founders finish in a day what used to take a year. If your process still assumes the old clock, someone else is already on the new one. Software is also cheap enough to build the exact tool you need for an audience of one. The old advice was scratch your own itch and hope a market appears. The better version is scratch it because the cost is close to zero. Some of those tools become companies when other people ask for them. The binding constraint is memory. Without pruning, the system is a junk drawer with excellent search. It will surface stale facts with total confidence. A bad skill file will run a bad process forever. Every fact needs a source. New information has to be checked against old information. Someone has to delete what no longer belongs. Treat that as infrastructure and the system compounds. Treat it as a dump and you get an agent that sounds certain and is wrong in ways you cannot audit.