Everyone sculpts in the repo

The old way had a lot of handoffs.

Each handoff is a translation. Nuance leaks at every seam.

The PM wrote the PRD in one tool. The designer interpreted it in another. The developer interpreted the design in a third. Each handoff lost nuance. The designer’s intent didn’t fully survive the trip to production. The PM’s constraints didn’t fully survive the trip to design.

Meetings, standups, handoff docs. These were the patches we built to stop context from bleeding out. They worked, mostly. But they were expensive, and they scaled with the number of seams, not with the quality of the work. The final product was a game of telephone.

One surface

Now put everything in one place. The PRD is a markdown file in the repo that both the human and the AI can read. The designer works from that PRD. The developer generating code has access to it too. The AI has all three contexts at once: the requirements, the emerging design direction, and the production code. The design prototype uses real components, sitting in a draft PR that the developer can further sculpt. Everything is open text files, versioned together, accessible to the same tools.

This is the shift that tools like Kiro and Cursor are building around: specs, requirements, and tasks live alongside the code. AI models are fluent in open text in a way they’re not fluent in proprietary file formats. They can read a markdown PRD, reason about a structured spec, generate a component, generate a design flow, and modify production code. A proprietary design file is a black box to an LLM. A repo full of plain text and code is a conversation.

Granola AI reads meeting notes from Granola and starts the next design iteration automatically. You steer, it builds.

Linear AI reads tickets, understands context, and can start work before you open your editor.

Design tools The traditional design tool shifts from source of truth to a refinement layer. MCP brings code in for visual tweaking and annotation.

Slack AI parses team feedback from threads and turns it into actionable iteration briefs.

AI is the connective tissue. Hover each spoke to see how.

AI is the connective tissue

The connective tissue that used to be meetings and handoff docs is now compute. AI reads the PRD, parses meeting notes, understands your intent, and acts on your behalf to iterate. It bridges the phases that used to require human translation. The compute does the heavy lifting through the loop. You steer.

The craft is the same

Systems thinking, visual hierarchy, information architecture, typography, interaction design. The skills that made a designer good in 2015 are the same skills that make a designer good now. Twenty years of training your eye doesn’t disappear because the surface shifted.

The designer’s job isn’t pixel-pushing. It never was, really. It’s judgment: what to build, how it should feel, what to cut. That judgment now gets expressed through steering AI agents instead of dragging rectangles.

Everyone sculpts in the repo. PM, designer, developer, agent. Different entry points, same material.