AI, systems, and product creation
AI should extend design, not flatten it.
I’m interested in the point where AI stops being a novelty and starts becoming part of a real product workflow. That means moving beyond isolated prompts and into structured systems that can support design system audits, documentation, design generation, code support, QA review, validation, and delivery in a coordinated way.
Practical futurism, no magic fog.
Human judgment stays central.
Systems create leverage.
Design philosophy
Useful AI has a job, a boundary, and a reviewer.
AI should create leverage, not noise
If a workflow produces more cleanup than progress, it is not useful.
Human judgment stays central
AI can propose, assemble, translate, and validate. Humans still decide what is right.
Quality control matters more than generation
The future is not just about generating more. It is about filtering better, validating better, and maintaining stronger system integrity.
Design systems are high-leverage AI surfaces
Documentation, governance, component maintenance, pattern validation, and design-to-code alignment are all strong candidates for structured AI support.
Agent workflow playground
Less single assistant, more coordinated product team.
I think of AI less as one assistant and more as a coordinated system of specialized agents with roles, handoffs, review loops, and governance.
Discovery Agent
- Where it helps
- Synthesizes inputs, customer pain points, competitive notes, and research fragments into structured opportunity areas.
- Do not trust alone
- Should not invent certainty or replace real customer evidence.
- Handoff
- Problem framing, assumptions, open questions, and research gaps.
Design System Agent
- Where it helps
- Checks components, tokens, naming, state coverage, and pattern reuse against system standards.
- Do not trust alone
- Should not silently rewrite system rules or approve its own recommendations.
- Handoff
- Component recommendations, drift warnings, and documentation updates for review.
Prototyping Agent
- Where it helps
- Turns a direction into testable flows, interface variants, or interaction models faster than a blank canvas.
- Do not trust alone
- Can optimize for surface polish before the concept is right.
- Handoff
- Prototype candidates, interaction notes, and unresolved decision points.
Documentation Agent
- Where it helps
- Keeps decisions, specs, usage guidance, and contribution notes alive as the work changes.
- Do not trust alone
- Documentation still needs human review for judgment, tone, and accuracy.
- Handoff
- Readable docs, changelog notes, and implementation guidance.
Code Agent
- Where it helps
- Translates structured design intent into implementation scaffolds, examples, or diffs.
- Do not trust alone
- Should be constrained by system rules, code review, and testable acceptance criteria.
- Handoff
- Code-ready output with assumptions, dependencies, and validation needs.
QA Agent
- Where it helps
- Checks output against accessibility, responsiveness, component rules, and known edge cases.
- Do not trust alone
- Can miss product nuance, context, or the difference between technically valid and actually good.
- Handoff
- Issues, severity, reproduction notes, and recommended fixes.
Governor Agent
- Where it helps
- Reviews proposed changes against product intent, design-system health, and risk.
- Do not trust alone
- Should support governance, not become an opaque approval machine.
- Handoff
- Approval notes, blocked items, and human-review checkpoints.
Workflow comparison
Design and code do not have to move in a strict sequence.
Traditional linear workflow
- Discovery happens, then fades into a deck.
- Design explores in parallel but documentation lags.
- Engineering receives specs late.
- Validation happens after expensive decisions.
- System cleanup becomes a future problem.
AI-enabled orchestrated workflow
- Discovery, docs, design, and code support stay connected.
- Agents create structured outputs with explicit assumptions.
- Design systems validate patterns earlier.
- Humans review decision quality, not just production volume.
- Feedback loops get shorter without removing craft.
Enterprise design systems + AI
Design systems are one of the strongest AI use cases.
Enterprise design systems involve repeated logic, documentation maintenance, component auditing, token management, governance checks, multi-platform translation, pattern consistency, and change management. That makes them a high-leverage surface for structured AI support.
Pattern drift3 components need reviewReview
Documentation health12 usage notes staleUpdate
Variable consistency2 token mismatchesValidate
Downstream impact5 files may breakTrace
Governance queue4 changes waitingApprove
Workflow toolkit
Tools are useful when the workflow has judgment built in.
My AI-assisted workflow exploration spans design, documentation, engineering handoff, validation, and governance. The tools change quickly; the important part is defining the job, the boundary, the review loop, and the human decision point.
- Figma
- FigJam
- Figma Make
- Claude
- Kiro
- ChatGPT
- Gemini
- GitHub
- GitLab
- Jira
- Confluence
- UserZoom
- Adobe Creative Suite
DAWin and personal experimentation
A proving ground for AI-enabled product development.
DAWin has been one of the strongest environments for exploring this space. Through that work, I’ve been thinking deeply about how AI can support end-to-end product development: from discovery and feature planning, through design, documentation, code generation, validation, testing, and deployment readiness.
It has also been a useful proving ground for the broader idea that design and code no longer need to happen in a strict sequence. With the right systems in place, they can move in parallel, creating faster feedback loops and getting ideas into realistic, testable form much earlier in the process.
What stays human
The more important the design decision, the more important human judgment becomes.
AI can accelerate the parts of product work that are repetitive, brittle, or too slow. But taste, ethics, product judgment, emotional nuance, and responsibility still sit with people. The goal is not automation for its own sake. The goal is better thinking, faster iteration, and more time spent on the work that actually requires human insight.