We’re building the design layer for the agentic era.
Creative work happens in many ways today. A designer works at a canvas. An agent generates a design on its own. A batch process resizes, localizes and rebrands one design into hundreds. All of this touches the same work, but our tools split it into separate systems.
Each system is built for one way of working. Design tools assume a person clicks every element. Generators produce a flat image you can look at but can’t precisely edit. Agents like Claude Code or Codex do great creative work, but they operate in an environment built for chat or development, not for design. MCPs and skills have exploded the ways we can be creative. Yet we get lost in trying, not in making.
On top of that, agent output feels duct-taped together, not built with intent. HTML as a rendering engine for jobs it was never meant to do. It doesn’t know what a page is, so print turns into workarounds: no bleed, no real CMYK workflow, no PDF/X. It doesn’t know what time is, so video turns into 900 screenshots. And what comes out is a flat file nobody can open and keep working on.
We need one layer underneath all the ways of working, not another tool for one of them.
We set out to build CoDesign with the following principles in mind:
- A real design document, not pixels and not HTML. Pages, layers, typed blocks for text, images, shapes and video. Templates with variables, constraints and locked elements. An engine that can deliver production grade print and video, and equally handle design documents like PDF, InDesign, PowerPoint or Photoshop.
- Equal access for humans and agents. Every operation a person can do on the canvas, an agent can do through an API or MCP, and the other way round. An agent’s output should be as editable as a human’s, and a human should be able to take over mid-stream without exporting and re-importing. Same model, different hands.
- Built for scale. Batch is a mode of the same engine, not a separate pipeline. You feed one design plus data and get hundreds of variants. Rendering is deterministic and headless, and brand rules hold in every variant.
- The output stays alive. What comes out is a file you can open and keep working on, not a flat end product. The editable source travels with the export.
- Intent as part of the format. Brand constraints, editable versus locked regions, and structure live in the document. The agent works within guardrails and hands you over not only a design, but an entire visual editor, tailored to that use case, so you can refine.
