The emerging paradigm is a central coding agent with multiple specialized input tools. A canvas tool (like Paper) will be for visual prompting, an IDE (like Cursor) will be for code refinement, and a text prompt will be for direct commands, all interoperating with the same agent to build software.
To enable AI-powered prototyping without production risks, large tech companies are creating separate, forked repositories for designers. This "designer playground" approach avoids the friction of production environments (e.g., linting, deploys) while providing a real-world starting point for stateful design exploration.
Contrary to claims that "handoff is dead," designers at top companies use AI-generated prototypes as highly detailed specs. These interactive prototypes provide more information than static designs but are still handed off to developers for implementation, rather than being merged directly into production.
A meta-workflow is emerging where designers use AI prompts not just to build the prototype, but to build tools *within* it. Examples include creating live version pickers for stakeholders or generating a markdown file that lists and controls all component states, effectively prompting a custom handoff tool.
Despite mandated adoption and new capabilities, there's no clear evidence yet that AI prototyping tools lead to faster production or better software. The time spent building a highly-detailed interactive prototype may not be quicker than traditional methods, and the complexity requires rigorous code review.
A flywheel effect is occurring: AI models excel with modern web stacks (Tailwind, Next.js), encouraging their adoption, which in turn improves the models. This will create a massive divide in workflows and capabilities between designers on modern stacks and those on legacy systems, making them almost different professions.
While forked codebases empower designers with AI tools, they create a new operational cost. Teams must now maintain two copies of the app—the real one and the designer one—which risks falling out of date. This mirrors the long-standing problem of syncing Figma design systems with production code.
In large companies, designers overwhelmingly use local AI coding tools (Cursor, Claude) over cloud-based ones (Replit, V0). The key advantage is using the company's real production app as a "starting place," which eliminates the need to recreate screens or components from scratch for every prototype.
Companies like Shopify and Atlassian now require designers to use AI tools like Cursor and Claude in their work, enforced through performance reviews. This top-down mandate aims to accelerate exploration of new workflows, such as stateful prototyping, and overcome the friction of adopting new tools amidst tight deadlines.
A key reason Figma won was its cloud-based, real-time collaboration. The trend of using local AI dev tools (like Cursor) is a step backward in this regard, reintroducing friction around sharing work and getting feedback, the very problems that led designers away from local files in the first place.