UX Collective: From vibe to specs with the SAID framework

Product teams increasingly use AI to generate interfaces and prototype ideas, but the handoff problem has gotten worse rather than better. Designs that start as vague “vibes” still have to become precise specifications, and the gap between the two has widened as generation speed outpaced articulation clarity. Darren Yeo’s piece in UX Collective addresses that gap with a structured workflow framework.

Yeo’s argument is that the traditional sequential process — from concept to wireframe to handoff to implementation — no longer suits teams where AI agents play an active role in building. When AI can execute from a spec, the specification becomes the design artifact. Static screens become less relevant; the behavior, constraints, and logic encoded in a document become more relevant. The piece makes a case for treating specs not as outputs from design but as the central shared workspace of a product team.

The SAID framework he introduces is a continuous loop with four stages. Specifying intent involves co-authoring specification-driven documents that define goals and constraints in a form readable by both humans and AI agents. Agentic delegation addresses the question of which tasks are suited to AI versus human judgment within that workflow. Iterative description covers the rapid cycling of design intent and prototype updates, replacing lengthy linear iteration. Discernment and diligence, the fourth stage, is where designers review AI outputs, evaluate quality, and apply governance before anything advances.

What makes this practically interesting is the shift in tooling it implies. Yeo describes Git as a potential shared workspace replacing the current chain of Miro, Figma, Confluence, Jira, and GitHub — a consolidation driven by specs becoming the source of truth rather than pixel-perfect screens. Designers who have spent years developing visual fluency are being asked to develop a parallel fluency in articulating behavior and constraints as structured, machine-readable text.

The article is most useful for product designers working on AI-native teams, engineering managers trying to align design and development workflows, and design operations leads figuring out where AI agents fit into their existing toolchain. It provides language and a framework for teams that are already experimenting with specification-driven development but have not yet formalized their approach.