AI product bottlenecks shift toward context, research quality and alignment

This week’s design discussions focused on what makes AI useful beyond model capability: accessible context, understandable limits and sound product decisions. Findings on Claude users and Miro’s research workflows exposed friction in maintaining context and deciding what to build as execution gets faster.

Key developments

Claude’s context files prove hard to manage

Claude power users often cannot find their context files and rarely edit them, according to usability findings from Nielsen Norman Group.

Context that users cannot find is harder to keep current, weakening their control over the information an agent relies on.

Oct 9 · Nielsen Norman Group

AI shifts research bottlenecks

Miro researcher Denis Kartepe says AI is speeding up research and blurring product-team roles, shifting bottlenecks from implementation to deciding what to build and aligning teams. He recommends structured documentation, human quality checks and direct engagement with users alongside AI-assisted workflows.

Faster execution puts more weight on problem definition and research verification in determining what gets built.

Oct 9 · UX Research Geeks

Also this week

Assistant design

• An analysis of the new Siri argues that useful assistants need relevant personal context, the ability to act on it and an experience that makes their limits understandable. Nielsen Norman Group

Product experimentation

• Josh Woodward, who leads Google Labs, the Gemini app and AI Studio, advocates frontier teams experimenting with new models, curiosity-driven side projects and links back to the core business to keep company labs from fizzling out. Lenny's Newsletter

What to watch

• Watch for a response to Claude’s context-file discoverability and maintenance problems; no decision or release timing is announced in the items.