Apply human-centered problem solving. Three frameworks dominate industry practice — pick based on what the user actually needs, don't default to IDEO's 5 phases.

When to Use

Ambiguous problem space, no obvious solution
"Why aren't users adopting X?" / "What should we build next?"
Stuck and needs structured divergence
"Who is this for?" / "Who should we target?" — audience definition and user persona synthesis
"Is this idea worth pursuing?" / "Should we invest in X?" — idea validation before committing resources
"We have too many ideas, which one?" / "Help me prioritize" — structured narrowing from many directions to one
"What problem are we actually solving?" — problem reframing when the real issue might be upstream
"Why did this product/feature fail?" — post-mortem through a design thinking lens (wrong problem, wrong audience, or wrong solution)

When NOT to Use

Known problem, known solution, just needs execution (skip straight to PRD — use product-manager skill)
Visual/UI implementation (use design skill)
Technical debugging
Pure information gathering with no decision needed (use deep-research skill)
Competitive positioning or feature comparison (use competitive-analysis skill)

Depth Calibration

Not every problem needs a full board. Before starting, gauge the scope and pick the right depth:

Quick Mode (10-15 min, 2-3 cards per phase)

Use when: the user has a specific question like "should I build X or Y?", the domain is narrow, or they just want a structured opinion. Research is 1-2 searches. Canvas output is compact — skip sidebar annotations, limit to 2 concepts, use a single assumption card. No forces diagram.

Standard Mode (30-45 min, full board)

Use when: the problem space is moderately open — "what should we build for market X?" or "why is feature Y underperforming?". Research is 3-5 searches across different angles. Full canvas board with all sections. Forces diagram for top concept only.

Deep Mode (60+ min, full board + extended research)

Use when: high-stakes strategic decisions, entering new markets, or pivoting a product. Research is exhaustive — 5+ searches, reading full articles via webFetch, cross-referencing data. Full canvas board with forces diagrams for every concept, detailed assumption mapping, and confidence scoring. Consider launching parallel research subagents for different angles.

Default to Standard Mode. Upgrade to Deep if the user's question involves a new market, significant investment, or a pivot. Downgrade to Quick if the question is binary or narrowly scoped.

Three Frameworks — Choose Wisely

1. Double Diamond (UK Design Council, 2004) — for ambiguous problems

Two diamonds = two diverge-then-converge cycles. Discover(go wide on problem research) →Define(narrow to problem statement) →Develop(go wide on solutions) →Deliver (narrow to one, ship it).

Critical insight: most teams skip the first diamond and jump straight to solution brainstorming. The first diamond exists to prevent solving the wrong problem beautifully.

Use when: problem isn't yet defined, timeline is weeks not days.

2. GV Design Sprint (Jake Knapp, Google Ventures, 2010) — for validation speed

5 days, 5 phases, hard time-box. Map (Mon) → Sketch (Tue) → Decide (Wed) → Prototype (Thu) → Test with 5 real users (Fri). Full method free at gv.com/sprintanddesignsprintkit.withgoogle.com.

Known weakness: users appear on Day 5, not Day 1. The sprint relies on team intuition to frame the problem — you can spend 5 days sprinting toward the wrong target. Fix: run JTBD interviews before the sprint to pick the problem.

Use when: problem is defined, team can commit 5 uninterrupted days, you need a go/no-go decision fast.

"5 users" is not arbitrary: Nielsen Norman Group research shows 5 users find ~85% of usability problems. Diminishing returns after that.

3. JTBD Switch Interview (Bob Moesta / Re-Wired Group) — for understanding why people buy

Not a workshop — a forensic interview technique. Interview people who recently switched (bought your product, or churned to a competitor). 45-60 minutes. Reconstruct their timeline, don't ask about features.

The Timeline (walk backward through their actual purchase):

First thought — when did it first occur to you this was a problem?
Passive looking — noticing solutions but not acting
Event one — something happens that makes it urgent
Active looking — comparing options, raised hand
Deciding — what tipped it?
Buying — the actual moment

The Four Forces (what made the switch happen):

| Force | Direction | Probe |

|---|---|---|

| Push of the situation | Toward switch | "What was happening that made the old way stop working?" |

| Pull of the new solution | Toward switch | "When you imagined having this, what got you excited?" |

| Anxiety of the new | Against switch | "What worried you about switching?" |

| Habit of the present | Against switch | "What was good enough about what you had?" |

Switch only happens when Push + Pull > Anxiety + Habit. If someone didn't buy, one of the blocking forces won.

Key technique: ask about the situation, never the feature. Not "do you like the dashboard?" but "walk me through the last time you opened it — what were you in the middle of?" Reference: the "Mattress Interview" at jobstobedone.org — Moesta interviews a guy about buying a mattress for 45 min and uncovers it was actually about his marriage.

Use when: you have customers but don't understand why they chose you, or you're losing deals and don't know why.

Research Synthesis

Research is only useful if it feeds directly into the framework phases. Follow this process to avoid the common failure of doing research and then ignoring it during ideation.

Step 1: Gather (DISCOVER phase)

Run 3-5 web searches across different angles of the problem space. For each search, extract:

Hard numbers — market size, penetration rates, growth figures, demographic data
Pain signals — what people complain about, what's broken, what's missing
Existing solutions — who's already solving parts of this, and where are they falling short

Step 2: Distill (DEFINE phase)

From the raw research, identify 2-3 key insights — non-obvious patterns that connect multiple data points. Each insight must be grounded in evidence from Step 1, not invented. Format each as:

Insight: [pattern observed]

Evidence: [specific data points or sources that support it]

Implication: [what this means for what we should build]

These insights directly feed the HMW statement. The HMW should address the most important insight, not the most obvious problem.

Step 3: Ground (DEVELOP phase)

Every solution concept must reference at least one insight from Step 2 and one data point from Step 1. This prevents "idea theater" — generating concepts that sound creative but ignore the research. If a concept can't be traced back to the research, cut it.

Step 4: Stress-test (DELIVER phase)

For the recommended concept, revisit the research to identify what could invalidate it. Look for counter-evidence, missing data, or assumptions the research didn't cover. These become the riskiest assumptions to test.

Core Tools (Framework-Agnostic)

How Might We (HMW) — reframe problem as opportunity. Scope test: too narrow bakes in the solution ("HMW add a share button"), too broad is unactionable ("HMW make users happy"). Right: "HMW help busy parents find 20-minute recipes without meal-planning guilt?"

Crazy 8s (from GV Sprint) — fold paper into 8 panels, sketch 8 distinct ideas in 8 minutes. Forces past the obvious first 3 ideas. Works solo or in groups.

Assumption Mapping — plot assumptions on Importance × Evidence 2×2. High-importance + low-evidence = test first. This picks your prototype target.

Empathy Map (Dave Gray, XPLANE) — Say / Think / Do / Feel quadrants around a user. "Think" and "Feel" are where insight lives — they're the gap between what users say and what they do.

5-Act Interview (GV test-day script): Friendly welcome → context questions → intro the prototype → tasks (watch, don't help) → debrief. One person interviews, team watches on video in another room and takes notes.

Adoption Forces (applied to every concept) — The JTBD Four Forces framework is too useful to reserve only for JTBD analyses. For every solution concept in DEVELOP, map these four forces from the target user's perspective:

| Force | Direction | Question to answer |

|---|---|---|

| Push | Toward adoption | What's painful enough about the status quo to make them try something new? |

| Pull | Toward adoption | What's exciting about this concept — what do they imagine getting? |

| Anxiety | Against adoption | What scares them about switching — cost, learning curve, risk, trust? |

| Habit | Against adoption | What's good enough about their current way — why haven't they already changed? |

Adoption happens when Push + Pull > Anxiety + Habit. If the forces don't balance toward adoption, the concept needs redesign — either amplify Push/Pull or reduce Anxiety/Habit. In Standard mode, do this for the top recommended concept. In Deep mode, do it for every concept.

Output Format


# [Challenge]

## Framework Used

[Double Diamond / GV Sprint / JTBD — and why this one]

## Problem Definition

### HMW Statement

### Key Insights (with evidence — quote or observation, not assumption)

## [If JTBD] Forces Diagram

| Push | Pull | Anxiety | Habit |

|---|---|---|---|

## Solution Concepts

| Concept | Desirable? | Feasible? | Viable? | Riskiest assumption |

|---|---|---|---|---|

## Adoption Forces (for recommended concept, or all concepts in Deep mode)

| Force | Assessment |

|---|---|

| Push (pain of status quo) | |

| Pull (appeal of new solution) | |

| Anxiety (fears about switching) | |

| Habit (comfort of current way) | |

| **Balance** | Push + Pull [>/<] Anxiety + Habit |

## Recommendation

### Conviction Level

[Low / Medium / High] — State your confidence and what would change it.

- **Current conviction**: [e.g., "Medium — the pain signal is strong but adoption behavior is unproven"]
- **Would increase to High if**: [e.g., "3/5 merchants in a pilot use the app daily for 2 weeks"]

- **Would decrease if**: [e.g., "merchants report they tried similar apps before and abandoned them"]

## Prototype Plan

- What to build (lowest fidelity that tests the assumption)
- Who to test with (5 users, recruited how)

- What "success" looks like

## Next Steps

Canvas Presentation

When presenting design thinking analysis on the canvas, follow these rules for a clear, skimmable board.

Shape Types & When to Use Each

| Shape | Use For | Why |

|---|---|---|

| note (sticky note) | Section headers, sidebar annotations, key stats | Fixed 200px width. Auto-sizes font based on text length — fewer words = bigger text. Perfect for bold single-word headers. |

| geo(rectangle) | Content cards, full-width banners, problem statements | Respectswandh— use for wide content (500px+ cards, 1600px+ banners). Setfill: "solid" and pick a color. |

| geo(arrow-down / arrow-right) | Flow indicators between sections | Shows the progression through the framework. Usefill: "solid". |

| text | Small inline labels like "vs" between forces | Auto-sizes, minimal visual weight. |

Layout Rules

Headers as sticky notes with minimal text — Use 1-2 words max (e.g., "DISCOVER", "DEFINE", "FORCES"). The note shape auto-scales font size inversely to text length, so short text = large, skimmable headers.
Content cards as geo rectangles — 500px wide for 3-column layouts, 780px for 2-column layouts, 1600px+ for full-width banners. Always set fill: "solid" and a color.
Left-column header notes, right-side content — Place header sticky notes at x=1400, content cards starting at x=1670 (leaving a 70px gap after the 200px-wide header note).
Vertical spacing — 80-100px between rows within a section, 120-150px between sections. This keeps the board breathable and skimmable.
Arrow shapes between sections — Place geoshapes witharrow-down geo centered horizontally between sections to show flow. Use the section's color.
Annotation stickies as sidebar — Place note shapes at x=3350 (far right) aligned vertically with their related section. Use for market stats, key quotes, risk callouts — short punchy content that adds context without cluttering the main flow.
Color coding by section — Be consistent:

Violet: title, concepts
Blue: discover
Green: define
Orange: develop / forces
Red: deliver
Yellow: key callouts (problem statement, switch rule, recommendation)
Light-* variants for content cards within each section

JTBD Forces layout — Use 2x2 grid: Push (top-left, light-red) vs Pull (top-right, light-green), Anxiety (bottom-left, orange) vs Habit (bottom-right, grey). Place arrow-right shapes and "vs" text labels between the pairs.

Canvas Gotchas

Sticky notes (notetype) are always 200px wide — thew parameter is ignored. Plan layouts around this constraint.
labelColor: "white" can cause rendering errors — avoid it. Use lighter fill colors where default black text is readable.
color: "black"on geo shapes withfill: "solid"causes errors — usegrey or a dark supported color instead.
Always call get_canvas_state before placing shapes — check for existing content and find empty space.
Delete before recreate — when updating shapes, delete first then create fresh. Updates to notes can behave unexpectedly.

Example: Placing a Section


1. Header note (big text):

{ type: "note", x: 1400, y: Y, w: 200, h: 200, color: "blue", text: "DISCOVER" }

2. Content cards (3-column):

{ type: "geo", geo: "rectangle", x: 1670, y: Y, w: 500, h: 260, color: "light-blue", fill: "solid", text: "..." }

{ type: "geo", geo: "rectangle", x: 2220, y: Y, w: 500, h: 260, color: "light-violet", fill: "solid", text: "..." }

{ type: "geo", geo: "rectangle", x: 2770, y: Y, w: 500, h: 260, color: "light-green", fill: "solid", text: "..." }

3. Flow arrow to next section:

{ type: "geo", geo: "arrow-down", x: 2300, y: Y+300, w: 80, h: 60, color: "blue", fill: "solid" }

4. Sidebar annotation:

{ type: "note", x: 3350, y: Y, w: 200, h: 200, color: "yellow", text: "$45.5B\nmarket by 2028" }

Research

webSearch("[user's problem domain] user research") for prior art
webFetch("https://www.gv.com/sprint/") for full sprint method
webSearch("[product category] reviews site:reddit.com") — unfiltered user language for push/anxiety forces
IDEO Design Kit free methods: designkit.org/methods

Hard Truths

Agent cannot run real interviews — can write the discussion guide, analyze transcripts the user pastes in, and build the synthesis
Personas without research are fiction. Push user for real data (support tickets, reviews, churn interviews) before building empathy maps
Most "design thinking" fails because teams do ideation theater and skip the uncomfortable research. Bias toward the first diamond