How to build a company when you're not an optimist | Jay Kreps (Co-founder and CEO, Confluent)

• A single 25-page blog post did more for Kafka adoption than years of engineering—the product marketing pyramid matters as much as the code
• Confluent bet everything on a cloud product when it was "embarrassing" and half the company opposed it, because they knew they "had to" succeed there to survive
• The 80% rule: CEOs need to know 80% of what their executives know about each function—enough to recognize good work, not enough to do it
• "What can we do?" vs. "What must we do?" is the most underrated building lever—once you know something is existential, you find a way
• As companies scale, they naturally become "Chipotle" (systematized mediocrity)—you need autonomous units with accountability to maintain pockets of excellence

Jay Kreps built Confluent from a scrappy group of engineers with zero go-to-market experience into a publicly traded enterprise software company, but not through typical founder optimism. His core insight: the difference between what a company can do and what it must do is one of the most underrated building levers. When Confluent's early cloud product was "embarrassing" and investors and half the company thought it was a terrible idea, Kreps pushed through anyway—because he knew they had to succeed in cloud to survive, even when the on-premise business was printing money at $100M ARR.

The journey started with a product marketing problem. Kafka had massive internal traction at LinkedIn but flopped as open source because no one understood what it was for. Kreps spent months writing a 25-page blog post that finally articulated the value—proving that product marketing requires as much rigor as engineering. That single post did more for adoption than years of development. His framework: build the full pyramid (evidence, use cases, detailed argument) before you can distill the top (the slogan). Most marketing fails because it's either all slogan with no substance, or all substance with no distillation.

On scaling, Kreps offers the 80% rule: CEOs need to know 80% of what their executives know about each function—enough to recognize good work and know if it's going well, but not enough to do the job yourself. You're operating in a "fog of partial understanding" and that's the job. The hardest part is diagnosing what's broken across functions, because everyone naturally blames other parts of the org or defaults to their own discipline's lens.

His most contrarian take: he's not an optimist, he's just tenacious. When you identify something the company must do to succeed, you stop asking "can we?" and start grinding until you find a way. This applies to hiring too—early sales reps need equity-heavy deals and experience at early-stage companies where there's no scaffolding, not Oracle veterans expecting a machine. As companies scale, they naturally become "Chipotle"—systematized and mediocre. The antidote is creating autonomous units with revenue targets and accountability, so they think end-to-end about customer success rather than just executing their narrow function.