In AI, it's easy to confuse "bigger" with "better." More parameters, more context, more tools, more eval scores often mean horizontal progress: scaling the same approach. Useful, but not automatically a product breakthrough.
Vertical progress is when new primitives change reliability and workflows: cheap inference, structured outputs, function calling, better retrieval, fast feedback loops, agentic automation, and deployment patterns that create real guarantees.
Builder implication: aim for systems that reduce uncertainty for the user. The win is "it can do this repeatedly with low variance," not "it can answer anything." That requires engineering: instrumentation, evals tied to user outcomes, guardrails, and tight product constraints.
A good heuristic: if your demo needs narration, you're likely doing horizontal progress. If your demo replaces a workflow without explanation, you're touching vertical progress.
A tiny vote of confidence
Was this note worth your time?