FORMAS Introduces CARTESIAN: AI Agents Inside Editable 3D Design

FORMAS has released CARTESIAN, a tool that places AI agents directly inside an editable 3D design environment. Announced via Parametric Architecture, the release targets a persistent gap in generative design workflows: most AI systems produce outputs that arrive as static geometry or images, leaving designers without a structured model they can query, revise, or rebuild.

CARTESIAN's core proposition is the pairing of two capabilities that design tools have historically kept separate. On one side sits the AI agent, which interprets a designer's intent and generates three-dimensional form. On the other sits the editable model — geometry a human can select, modify, and re-parameterize after the machine has produced it.

This distinction matters for professional practice. Architectural and product design teams evaluating generative tools consistently hit the same wall: an output that cannot be edited is a rendering, not a design. A parametric model carries relationships between elements — a facade module tied to a grid, a structural depth tied to a span — and those relationships are what let a team iterate through revisions without rebuilding from scratch. By keeping the AI's output inside an editable structure, CARTESIAN aims to convert generative output from a conversation piece into a working artifact.

The name signals the tool's underlying logic. Cartesian geometry — positioning every element in explicit coordinate space — provides the framework in which the AI agents operate. Rather than generating free-form approximations, the agents work within a structured spatial system, which is what makes downstream editing feasible. Each generated element has a definable position, relationship, and parameter set.

For interface watchers, the release fits a broader pattern. Design software has spent three years absorbing generative AI at the edges of the workflow: text-to-image concept boards, background removal, material suggestion. The harder integration problem — putting an agent inside the actual authoring environment, where its output must survive contact with revision history, versioning, and collaboration — has seen far fewer credible attempts. Tools like CARTESIAN are a test of whether agentic AI can operate as a participant in an editable model rather than a generator adjacent to it.

The workflow change is concrete. Instead of prompting an external system, exporting a result, and importing it into a modeling package for cleanup, the designer stays in one environment. The agent proposes; the designer edits the proposal directly, adjusting parameters and geometry; the agent responds to the modified model. The loop closes without file translation, which is where format fidelity and parametric relationships typically break in current pipelines.

FORMAS positions CARTESIAN for parametric architecture audiences in particular, where the tension between generative speed and editability is most acute. Architectural teams generate dozens of massing and facade options in early phases, then spend significant effort reconstructing the chosen option as a proper parametric model. A tool that generates editable geometry from the start removes that reconstruction step.

Questions remain that the announcement does not address: pricing, platform availability, export formats, and how the agents handle revision at scale — whether an edit to one parameter propagates through agent-generated geometry as predictably as it does through human-authored parametric models. Those details will determine whether CARTESIAN lands as a production tool or a promising demo.

What is clear is the direction. Editable 3D output, not image generation, is the current frontier for AI in design tooling, and FORMAS has staked its claim with a release that treats the agent and the model as a single system. Expect the next wave of competitors to answer on the same ground.