Ep 22: How to Build Category-Defining AI Companies w/ Gokul Rajaram (DoorDash, Square, Google)
The "Compound or Die" Playbook & The Death of Middleware
In our latest episode of the UNPAK3D Podcast, we sat down with Gokul Rajaram—prolific angel investor, board member, and former executive at Square, Google, and Facebook—to discuss what separates winning AI companies from the noise.
The Three Capabilities Reshaping Software
Gokul sees three fundamental innovations opening new possibilities in AI:
Document processing that can now reliably extract insights from medical records, legal contracts, and complex unstructured data
Browser automation replacing legacy RPA tools to navigate business-critical workflows
Voice and conversational agents enabling tactical, multi-turn conversations at scale
Companies combining one or more of these capabilities at the application layer are where Gokul is spending his time—and notably avoiding infrastructure plays due to “leapfrog risk” as models continue eating into that layer.
The New Defensibility Playbook
With horizontal platforms like Google building their own agent composers, how do application companies survive? Gokul’s answer: proprietary reward functions.
You have to build a domain-specific evaluation system and reinforcement learning mechanism tuned to the specific workflows in your target market. The goal is creating self-improving systems with custom data that make it not worth a horizontal player’s time to compete in your niche.
Compound or Die
Perhaps the most striking shift: the old rule of waiting until $10-20M ARR before building product two is dead. The time to build your second product is as soon as your first product has product-market fit. Customer relationship is one of the biggest moats in software today. You want to surround the customer with as many interlocking, connected products as possible.
He recently led a Series A in a company with six revenue-generating products—something unheard of just a few years ago.
Retention Over Growth
When evaluating companies, Gokul cares less about top-line growth than most investors. His focus? Gross and net revenue retention. If you have very high NRR, it means customers are growing. But it has to be balanced with GRR—customers sticking.
Usage data matters even more than revenue data, he noted, because usage is a precursor to future expansion.
The 996 Reality
Finally, Gokul was candid about the intensity required to win. AI-native companies are operating in “managed chaos”—minimal product managers, everyone expected to build, and 996 schedules (9am-9pm, six days a week) becoming standard.
The notion of work-life balance is overstated early in your career if you want to build something remarkable, he said. “We don’t realize how much effort Bezos, Elon, or Zuck spent.”