Component specs, token specs, and contracts are converging on the same idea. Let the model do the heavy lift once, then hand the work to machinery that can't improvise.
Hey! I love the direction y'all are moving in. Looking forward to seeing how things progress. And thanks for the FCMCP support!! I love hearing from folks like yourselves that are using it for the power lifting 💪
Having deterministic gates really eliminates that endless state drift in agent loops. We've been exploring similar contract-driven design and workflow strategies over at Clickel Digital and it’s encouraging to see the whole ecosystem converging on standard specs like DTCG and DSCS. Thanks for compiling such a solid list of tools!
This maps exactly to something we keep relearning while building a design-system template with agents: prompt history is not project state. The transcript swells endlessly, but the agent builds from the files on disk — so the real risk was never the number of prompts, it's state drift. Which suggests a working rule that mirrors your authoring/enforcement split: anything you manage to pour into code, tokens, or a template stops drifting through the prompt. Every decision that stays conversational remains negotiable — the model can be talked out of it tomorrow. Every decision that becomes a file is settled until deliberately changed. So we measure progress less by what the agent got right today, and more by how much of yesterday's judgment no longer needs the agent at all.
Founder of Aiko here. Thanks for the shoutout!
Really enjoy your articles and Figma Console MCP is the tool of choice for us to read/write anything agentic to Figma.
Hey! I love the direction y'all are moving in. Looking forward to seeing how things progress. And thanks for the FCMCP support!! I love hearing from folks like yourselves that are using it for the power lifting 💪
Having deterministic gates really eliminates that endless state drift in agent loops. We've been exploring similar contract-driven design and workflow strategies over at Clickel Digital and it’s encouraging to see the whole ecosystem converging on standard specs like DTCG and DSCS. Thanks for compiling such a solid list of tools!
This maps exactly to something we keep relearning while building a design-system template with agents: prompt history is not project state. The transcript swells endlessly, but the agent builds from the files on disk — so the real risk was never the number of prompts, it's state drift. Which suggests a working rule that mirrors your authoring/enforcement split: anything you manage to pour into code, tokens, or a template stops drifting through the prompt. Every decision that stays conversational remains negotiable — the model can be talked out of it tomorrow. Every decision that becomes a file is settled until deliberately changed. So we measure progress less by what the agent got right today, and more by how much of yesterday's judgment no longer needs the agent at all.