Suleiman Shakir: Using AI as a design thinking partner with structured context

Suleiman Shakir is a product designer who has worked at Amazon, BuzzFeed, Siemens, and Handshake. In a May 2026 post on his personal blog — also published on UX Collective — he describes a system he built to use AI not as a visual generation tool but as a context-holding thinking partner across long design projects.

The problem he was solving. Most design projects involve accumulating large amounts of context: prior research, past decisions, failed directions, feedback cycles, and design constraints. In practice, that context gets stored across documents, Figma files, Slack threads, and individual memory. Reconnecting with it at the start of each session takes time, and things get lost. Shakir wanted the AI to hold that context so he could think about the problem rather than reconstruct the information.

The system. He built a structured project folder that Claude Code reads at the start of each session. The folder contains four files and directories: MEMORY.md, a running index of all decisions and data collected over the project’s life; CLAUDE.md, project-specific instructions for how the AI should behave; and separate data/, design/, and project-context/ directories for research, metrics, and user feedback.

Alongside this, he defined a set of custom Claude skills for recurring workflows: /explore for full design iteration from problem to prototype concepts, /brainstorm for early direction exploration, /synthesize for integrating new data into existing decisions, and /design-partner for pressure-testing a proposed solution against the original problem statement. He also uses MCP to connect Claude to Linear, Figma, and Slack, so the AI can pull live context without requiring manual updates.

The ai-design-buddy tool. To help others try the approach, Shakir released an open-source starter kit on GitHub called “ai-design-buddy.” It comes pre-loaded with a fictional local discovery app called Beacon, including sample research, personas, and custom skills. Users can run the /explore command to see a full design iteration, then adapt the structure for their own projects.

What this approach changes. In Shakir’s practice, the main effect is eliminating context-switching overhead across sessions. Decisions compound over time because nothing gets lost. Code prototypes generated by the AI also shift the feedback dynamic: reviewers respond to behavior and flows rather than to visual polish, which surfaces more useful UX feedback earlier.

What it does not replace. Shakir is explicit that this system does not substitute for design judgment. His note: “I’ve never shipped a direction without reshaping it because often times its suggestions can be quite terrible.” He describes the division as roughly 80/20 — the AI does heavy lifting on structure and synthesis, but the designer redirects and finishes the work. The system can also amplify existing blind spots in research, so its output is only as good as the inputs.

Who it is useful for. Product designers working on projects that extend over weeks or months, where context accumulation is a real problem. Less useful for short-duration tasks where the project is simple enough to hold in working memory. Requires upfront investment in organizing the folder structure, and some comfort with Claude Code and terminal-based workflows.