Section 03

Design Thinking

juniormidsenior

Overview

Design thinking is a human-centred, iterative methodology for solving complex problems. Its five phases - Empathize, Define, Ideate, Prototype, Test - are not a linear process; they are a thinking mode. AI accelerates execution within each phase but is structurally incapable of doing the one thing design thinking requires: genuine empathy with real humans. Use AI to move faster inside a framework you understand deeply. Do not use AI to skip the empathy phase.

AI workflow

8 steps
  1. 01

    Empathize - Research Planning

    Claude + QoQo

    Define who you are empathising with and what you need to understand. Use QoQo to generate initial proto-personas. Use Claude to identify what secondary research already exists and what gaps require primary research. Build empathy maps before writing any interview questions.

  2. 02

    Empathize - Field Research

    Otter.ai + Loom

    Conduct 5-8 user interviews. Use Otter.ai for real-time transcription (300 min/month free). For fixed-bid projects, use Loom for async research (25 free videos). The goal: understand context, mental models, jobs-to-be-done, and emotional friction - things AI cannot simulate.

  3. 03

    Define - Insight Synthesis

    Claude + Miro AI

    Paste anonymised interview themes into Claude: "Synthesize these research insights into a clear problem statement using the format: [User type] needs [need] because [insight]. Generate 5 How Might We questions at different scales of ambition." Use Miro AI to cluster themes visually.

  4. 04

    Define - Problem Statement

    Claude

    A good problem statement is specific, human-centred, and leaves solution space open. Prompt: "Here is our current problem statement. Critique it: Is it too solution-oriented? Is it specific enough? Does it reflect a real user need or a business assumption? Rewrite it with these corrections."

  5. 05

    Ideate - Divergence First

    Human session + Claude

    Human-only ideation for 25 minutes before using AI. Then use Claude as a provocation engine: "Apply extreme constraints to this concept. What if it were free? What if it worked offline? What if it were designed for 80-year-olds? Push toward unexpected territory."

  6. 06

    Prototype - Fast and Purposeful

    v0 / Figma / Google Stitch

    Prototype to answer a specific question, not to impress. Define the learning goal before building. Use v0 or Google Stitch for 30-minute lo-fi scaffolding. Use Figma for higher fidelity when testing visual or interaction-specific hypotheses. Match fidelity to the question.

  7. 07

    Test - Plan and Execute

    Maze + Claude

    Generate test plans with Claude: "Here is the prototype and the design hypothesis. Write a 5-task usability test with success criteria for each." Run with Maze for unmoderated testing or moderated sessions with Loom. Synthesize findings back into the Define phase.

  8. 08

    Iterate - Close the Loop

    Claude + Notion

    After each test cycle: "Here are the usability findings from this round. What did we confirm, what did we disconfirm, and what new questions emerged? What is the highest-priority change for the next iteration?" Document the iteration rationale - this is what makes design work teachable.

Tools

ToolUseFree tierClient-safe
ClaudeSynthesis, HMW generation, ideation provocation, iteration planningFree tier availableYes
Miro AIAffinity mapping, opportunity clustering, visual synthesis3 boards freeYes
Otter.aiEmpathy phase interview transcription300 min/month freeYes
v0 by VercelRapid prototype scaffoldingFree tier availableYes
MazePrototype testing with AI summariesLimited free responsesYes
QoQoPersona generation, user journey mappingFree tier availableYes

Constraints

4 constraints
  • Design thinking is a mindset, not a checklist. The phases are a scaffold for thinking, not a waterfall process. Revisiting earlier phases mid-project is correct behaviour, not failure.
  • AI cannot empathise. The Empathize phase requires real humans. Synthetic users (AI-simulated personas) are useful for early directional validation but are not a substitute for the discovery that comes from speaking with real people.
  • Fixed Bid: Compress each phase to its minimum viable output. AI can halve the time of Define, Ideate, and Prototype phases. Empathize and Test phases require human time that AI cannot compress.
  • Consultancy: Document AI-assisted outputs with the human judgment layer visible. Show clients the raw data, the AI synthesis, and the human interpretation as separate artefacts.