Questions readers send us
July 11, 2026 edition- Where should a team actually start?
- In this order: exhaust the native layer (Figma AI and its linter), add one enforcement gate (the Francoise or FigSpecs class), document the system properly (the zeroheight class), and only then point AI at code (the Builder.io class). Teams that skip the documentation step discover that agents improvise, and improvising agents are worse than none.
- At what point do tools stop being the answer?
- When your designers spend more time shuttling between AI tools than designing. Five disconnected assistants still leave humans doing the glue work. That is the signal to buy orchestration as a service - external if the problem is volume (the Superside shape), embedded if the problem is the pipeline itself (the Humbleteam shape).
- How do we keep model output on-system?
- Every working setup we have reviewed rests on the same three legs: a design system documented as rules rather than folklore, automated compliance checks that run without being asked, and one accountable human sign-off before anything reaches customers. Figma's 2026 survey found only about a third of designers trust raw model output; the three legs are what the other two-thirds are missing.
- Is a directory like this ever neutral?
- No publication that selects is neutral, which is why we publish the yardsticks and link every entry to its source. Nothing here is paid for. Vendor-commissioned research is identified as such wherever we cite it. Read the selection notes and check our work - that is what the links are for.
- Why are Midjourney and Canva not listed?
- Because they serve a different reader. Generation tools for individual creators are excellent products answering a question this directory does not ask. The entries here are chosen for multi-designer teams operating a design system on a revenue product.