---
title: "AI Design Practice | Parmis Meshgi"
description: "How senior product designer Parmis Meshgi uses AI for research synthesis, concept exploration, design systems, prototypes, code, and decision records."
canonical_url: "https://meshgi.com/ai-design-practice/"
html_url: "https://meshgi.com/ai-design-practice/"
author: "Parmis Meshgi"
location: "Toronto, Canada"
language: "en-CA"
last_updated: "2026-09-29"
---

> How senior product designer Parmis Meshgi uses AI for research synthesis, concept exploration, design systems, prototypes, code, and decision records.

Canonical page: [https://meshgi.com/ai-design-practice/](https://meshgi.com/ai-design-practice/)

[← Hiring guide](https://meshgi.com/restaurant-product-design/)

AI design practice

# More context.
Less repeated work.

I use AI to read the real sources, test more directions, build working prototypes, and keep decisions available after a project ends.

Working rule AI does not invent the product.

It uses verified requirements, current screens, known components, and review from people who own the decision.

The working loop

## AI has a job at each useful step.

The designer still owns the question, evidence, choice, and final review.

- 01

### Read the sources

Connected Figma, Jira, meeting records, product documentation, and code give the work current context.

- 02

### Synthesize research

AI groups interview notes, repeated pain points, constraints, and open questions. I check the source before using a claim.

- 03

### Explore more than one answer

I ask for distinct concepts, failure states, and tradeoffs. Weak directions end before the team spends time polishing them.

- 04

### Use the real system

Instructions name current components, tokens, flows, and product rules. New patterns require a clear gap and team review.

- 05

### Build and test behavior

AI-assisted code turns selected flows into interactive prototypes. I test responsive behavior, states, keyboard use, and recovery.

- 06

### Keep the decision

Finished work becomes a searchable record with the reason, source, owner, limits, and release state.

What this changes

## AI helps the team spend time on decisions.

Research

### Faster synthesis

Interview notes, product history, and connected stories become a source-linked brief before design starts.

Concepts

### More real alternatives

The team can compare several behaviors and their edge cases before committing to a visual direction.

Prototype

### Working interaction

The Kiosk case includes a coded flow for service type, menu, options, order, payment, and recovery.

[Try the Kiosk proof →](https://meshgi.com/kiosk/#product)

System

### Current patterns

Product instructions stop each new task from restarting with guessed components or old behavior.

Code

### Closer to implementation

I use AI-assisted code to test interaction and responsive rules. Engineers still own product code review and release.

Memory

### Decisions stay searchable

The next project can reuse the reason behind a rule instead of copying a screen without its context.

Guardrails

## Speed only helps when the work stays true.

Source before claim

Important facts link back to product, research, or release evidence.

Existing component first

The system stays consistent unless the product has a verified component gap.

Human decision owner

AI can compare options. The responsible designer, product manager, or engineer approves the decision.

Release state stays visible

Shipped, partial, proposed, and unmeasured work keep separate labels.

Recorded practice

## 7 platforms studied.
47 design versions.
200 users tested.

These numbers come from my April 2026 design Town Hall. They describe the work behind one polished screen.

[Return to hiring guide →](https://meshgi.com/restaurant-product-design/)