Founders should focus on their unique knowledge and interests rather than pursuing trendy markets. Consensus is often wrong, and the best opportunities lie in contrarian bets based on direct experience, a lesson Garry Tan learned after abandoning web development right before Web 2.0.
Garry Tan recounts turning down a founding role at Palantir to chase a promotion at Microsoft. The key lesson is to prioritize working with brilliant people over conventional career progression, trusting direct experience ('the territory') over the consensus 'map.'
Truly innovative companies often start with a small group sharing a strong, non-consensus belief. This 'cult-like' culture, built on a secret truth that flies in the face of orthodoxy, is a feature—not a bug—of world-changing startups.
While co-founders were once a near-requirement for startups, AI tools and agentic coding now allow a single individual to achieve the productivity of a large team. This makes ambitious solo-founder ventures far more feasible than ever before.
The traditional per-seat SaaS model is becoming obsolete due to AI. Startups must now see it only as an initial wedge to build a deeper, more defensible moat around proprietary data, network effects, or another unique advantage.
By codifying a task into a 'skill file'—a combination of markdown instructions, code, and tests—companies can create AI-powered 'employees' that execute processes flawlessly and repeatedly. This transforms one-time human effort into a permanent, scalable asset.
AI agents can synthesize vast amounts of internal communication, like meeting transcripts, to give leaders unprecedented visibility into their organizations. This allows them to identify latent conflicts, understand ground truth, and make perfectly timed, high-context interventions.
The Toyota Production System succeeded by empowering line workers—those with the most context—to improve the process. Similarly, AI systems will improve fastest when the 'agents' doing the work can provide feedback and directly influence the system's design and processes.
Fears of rapid, mass job displacement from AI are tempered by a powerful counterforce: institutional inertia. The bureaucracy and human limitations in every large company and government will act as a natural brake, slowing AI adoption and giving society decades to adapt.