Product designer Getter Chen highlights a new AI workflow centred on specialised tools, open standards, and durable knowledge bases that streamline tasks and enhance model efficiency in design processes.
The emerging AI workflow described by product designer Getter Chen is less about finding a single model that can do everything and more about building a balanced system of specialised tools. Chen compares the process to deck-building games such as Balatro: individual cards may look powerful on their own, but the real advantage comes from combining them in the right sequence for the task at hand.
At the centre of that system is the Model Context Protocol, or MCP, which Chen treats as a support layer rather than a primary engine. Figma describes MCP as an open standard introduced by Anthropic in November 2024 to connect AI systems with external tools and data sources through a common interface. In practice, that makes it useful for reducing manual switching between read and write tasks, but also expensive in tokens, the computing budget that determines how much work a model can carry out before costs rise or context runs out.
Chen’s second pillar is what he calls an LLM Wiki, a way of recycling past output into a structured knowledge base that future prompts can reuse. The concept was popularised by Andrej Karpathy in April 2026, when he described a filesystem-based memory system designed to give models durable recall beyond a single context window. Reporting and community write-ups say the approach has been adopted widely, with some implementations growing into hundreds of articles and hundreds of thousands of words.
That emphasis on durable memory is important because it changes where the model does its most valuable work. Rather than treating each conversation as a fresh start, the wiki lets earlier research, notes and drafts become a reusable layer. Aaron Fulkerson, writing about a production version of the pattern, said the method can sit between raw inputs and active agents, turning scattered material into something that is easier to query, refine and extend over time.
Chen’s main production tools remain Codex, ChatGPT and NotebookLM, which he says feed material into the wiki, while dedicated skills are used to keep critique grounded and less flattering. For design work, he sees Figma Agent as the most practical arena for review and direct text-to-interface generation because it keeps the work close to the canvas and lowers the cost of iteration. The result is a workflow built around output efficiency: MCP to reduce friction, LLM Wiki to preserve and compound value, and AI models to generate the work that feeds the next round.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
