Co-Designing Products with AI

Most builders open a session with a feature list: "Build me a language learning app with flashcards, spaced repetition, and grammar exercises." This produces a competent remix of existing apps. The AI has seen thousands of language learning specs and will generate a polished version of what already exists.

A co-design session opens differently. "People just want to make themselves understood. Not learn grammar. As long as you make yourself understood, everything else is just a bonus." This is vision-first design — starting from the human experience you want to create. The AI structures around the pragmatic goal, not a feature checklist. The product that emerges looks fundamentally different: real-world conversations, practical daily scenarios, and connections with actual people — the woman in the bakery, the cashier in the supermarket, the old woman with the dog — instead of flashcard decks and grammar drills.

The second pattern reinforces the first. After forming your vision, use AI research to validate it — not to generate your direction. Ask: "Does research support the idea that communication-focused language learning — where the goal is being understood rather than grammatical accuracy — increases long-term retention?" This grounds your instinct in evidence without letting the evidence constrain the vision.

Builders who start with research often end up building what the research describes. Builders who start with vision and then validate with research build something new that happens to be supported by evidence.

Trying to specify everything upfront produces either a shallow document or an exhausting marathon session. Iterative deepening works in expanding circles:

• Layer 1 — Core concept. The emotional truth. Why should this exist? What experience are you creating?
• Layer 2 — Key mechanisms. How does it actually work? What are the moving parts?
• Layer 3 — Engagement model. Why do people come back? What creates the habit loop?
• Layer 4 — Edge cases and system design. What happens when things go wrong? How does it scale?
• Layer 5 — Gap analysis and decisions. What is missing? What trade-offs need resolving?

Each layer builds on the previous one. The AI remembers the full context and connects new ideas back to earlier layers. You can stop at any layer and have something coherent. This is the design equivalent of progressive enhancement.

During this process, rough input is your best tool. Half-formed thoughts, stream-of-consciousness messages, even typo-laden fragments — these are creative seeds. Twenty unpolished words can become a two-thousand-word architectural section. The AI's job is to grow the seed. Time spent polishing your input is time not spent generating ideas.

When the AI's output misses something, extend rather than reject. Instead of "that section is wrong," say "I think we should also consider daily challenges and social sharing." Correction by addition keeps the creative flow going. Rejection triggers re-generation and discards existing work. Addition triggers extension and preserves it.

Co-design has two distinct cognitive modes. Expansive mode is creative building — adding, extending, imagining. Contractive mode is auditing — checking completeness, finding gaps, making decisions. Trying to do both simultaneously produces neither good ideas nor thorough reviews.

After the building phase, make the shift deliberately. Ask the AI: "What's missing? What do we need to sort out?" This is gap analysis as a deliberate step. The AI has read every word of your design document and can systematically compare sections against each other. It catches inconsistencies and omissions that you cannot see because you wrote the content in layers over time.

AI never forgets what you have already written. That makes it the best gap-finder you will ever work with.

When gap analysis surfaces ten open questions, a builder with a clear vision answers them all in one message. "Standard language only for version one." "Into action as soon as possible." "No net cost increase." These decisions flow instantly from clear principles. If gap decisions feel hard, the vision is not clear enough. Go back to layer one and sharpen it. Slow decision-making during design is almost always a symptom of unclear principles, not insufficient information.