Stephen Haney: the 2026 AI design field report — what's actually working

Published on January 19, 2026, this video presents findings from Stephen Haney’s field research into how professional design teams at companies like Shopify and Notion are integrating AI into their actual work. Haney is the founder of Paper, a design tool, and conducted the research by interviewing practicing designers at organizations with established AI workflows rather than those still in the exploration phase.

The video is aimed at designers who want to understand the current reality of AI adoption inside product teams—specifically what tools are being used, how workflows have changed, and what friction has emerged.

Key takeaways:

1. Claude Code and Cursor have become the preferred tools for AI-assisted design work. Rather than standalone prototyping platforms like v0 or Replit, teams at these companies are using code-capable AI tools to prototype from existing production codebases. This means designers are working in environments close to the real product rather than isolated sandboxes, which produces prototypes with more accurate component behavior and visual fidelity.

2. AI tool proficiency is now part of performance evaluation at leading companies. Adoption has moved from optional to expected. Several teams Haney interviewed had formalized AI tool use in their review frameworks, which changes the incentive structure for designers who might otherwise delay integration.

3. The collaboration gap is a real problem. Code-based prototypes are harder to share and gather async feedback on than Figma files. Haney describes this as a step backward compared to the pre-agent era: “we’ve actually taken a step backward in how we share work.” The problem is comparable to the collaboration challenge before Figma made multiplayer design standard.

4. Infrastructure is a bottleneck. Teams fork production environments to give designers a playground, but designers frequently struggle with the operational side—environment variables, databases, linting, and deployment processes that engineers handle automatically.

5. The startup-enterprise divide is widening. Early-stage product designers already write code and ship pull requests as part of their routine. At established companies, the gap between what is technically possible and what operational processes support is much wider.

Worth watching if you are responsible for how a design team adopts AI tools, are curious how leading product companies have actually changed their workflows (not what they plan to change), or want to understand the new bottlenecks that come after AI tool adoption—particularly around prototype sharing and infrastructure access.