# Design Thinking in AI4SE: Rhythm from the Double Diamond, Shape from Templates

With the same Superpowers brainstorming skill, one person ends a session with scattered fragments; another walks away with a consensus elevator pitch and a structured user journey map. The gap is rarely the model. It is whether Design Thinking rhythm, classic template shape, and an AI brainstorming engine are stacked together.

This site already covered why Clarify beats Prompt when brainstorming with AI—diverge–converge, problem vs solution space, and human-owned decisions. This post does not rehash that operating discipline. It adds the missing piece: in software requirements analysis, how the Double Diamond sets rhythm and templates set shape so AI brainstorming becomes both precise and fast.

1. Three-layer stack: rhythm · shape · engine

Treat effective human–AI requirements work as a three-layer stack:

Layer	Question it answers	Typical carrier	Failure mode if missing
Rhythm	Diverge or converge? Problem or solution space?	Design Council Double Diamond	Tech choices before problem definition; or endless unfocused diverge
Shape	What “done enough to accept” looks like this round?	Elevator pitch, empathy map, journey map, story map, service blueprint, etc.	Lots of AI prose; still no shared definition of “clear”
Engine	Who expands options, probes ambiguity, structures alternatives?	Superpowers-style AI brainstorming skills / flows	Purely human brainstorm limited by room memory and pace

Each layer owns one job:

• Double Diamond owns rhythm. The first diamond (Discover / Define) lives mostly in problem space—understand users, define the challenge. The second (Develop / Deliver) enters solution space—explore delivery options and filter. Understanding the model helps you define problems more precisely and find solutions more efficiently.
• Templates own shape. Classic frameworks are not Mad Libs; they supply exit criteria for convergence—e.g. an elevator pitch must answer who, what pain, and how you differ from the alternative. Clear shape gives multiple agents, people, or rounds a shared language.
• AI brainstorm owns the engine. It is strong at paralleling framings, probing blind spots, and turning rough ideas into candidate sets. Trade-offs at convergence stay human—when the AI is unsure, that is exactly when people should decide.

In one line: rhythm tells you which diamond you are in; templates tell you what convergence looks like; AI helps you run the diamond.

2. Playbook A: online—rough idea → multiple elevator pitches → consensus

Requirements often start as “I roughly know what I want, but I cannot say it cleanly.” That is normal. You do not need a complete PRD on day one.

A reusable online playbook:

1. Diverge in problem space. Feed the AI a rough idea—who the user is, what hurts, what success might look like. Use a brainstorming skill to explore problem definition; deliberately defer tech stack talk.

2. Converge into a template shape. Slot the discussion into an elevator-pitch structure. Moore-style positioning (from Crossing the Chasm) gives fixed slots, for example:

   For [target user] who [pain / need], [product] is a [category] that [key benefit]. Unlike [primary alternative], we [differentiation].

3. Generate multiple drafts, then converge again. Steer different agents (or multiple rounds / framings of the same skill) to each produce an elevator pitch. You will get several similarly structured, differently emphasized candidates—that is useful. Run another brainstorm pass that explicitly compares and converges the options; humans decide; you leave with one consensus pitch.

Operating rules:

• Exit criteria for round one: “at least 2–3 parallel elevator pitches,” not “chat until the AI says we are done.”
• Exit criteria for round two: “one consensus draft + rejected options and why”—otherwise you only have fake convergence.
• Only after the pitch is written down should you enter solution space: scope slices, story mapping, or Spec / Plan (see AI Development Workflows Landscape).

The template’s value is not prettier copy. It turns “what is the product?” from chat tone into a structure that is alignable, challengeable, and iterable.

3. Playbook B: offline whiteboard + audio/photos → AI re-convergence

Not every insight belongs in a chat window first. Journey maps and empathy maps often work better when people diverge and converge together at a whiteboard—bodies, sticky notes, and argument are part of alignment.

To wire an offline workshop into the AI engine:

1. Record audio (with consent). Transcripts after the session are high-value input for a second AI pass; photos alone lose most of the dialogue.
2. Complete one human diverge–converge cycle on the template—e.g. empathy-map quadrants, or journey stages–touchpoints–pains–opportunities. Reach a “current version” in the room, tidy the board, then photograph it.
3. Feed transcript + photos to the AI. Use brainstorming to structure, find gaps, and propose candidate convergences—“Are these two pains the same? Did this opportunity jump into solution space?”
4. Humans decide where the AI is unsure. The model will surface ambiguities. The response is not “let AI pick,” but human judgment and constraints written back into the document.

The same pattern works for user story mapping: lay the backbone and slices offline, then audio + photos into AI to refine release slices and open questions. For how to staff workshops and funnel decisions, see AI4SE workshop design strategy. This post only stresses the interface: whiteboard artifacts must be consumable by AI for a second pass, and that second convergence is still human-owned.

4. Payoff, boundaries, and anti-patterns

The main payoff of the whole path: when you understand Design Thinking, borrow classic templates (elevator pitch, journey maps, story maps, service blueprints), and use AI brainstorming professionally, speed and quality rise together. The Double Diamond keeps you from being diligent in the wrong space; templates keep convergence from being shapeless; AI keeps diverge-and-structure from depending only on room memory.

Common failure modes:

Anti-pattern	Looks like	Fix
Skipping a diamond	Architecture / stack talk before the problem is defined	Until the first diamond ends, artifacts stay at problem statement and success criteria
Template without decision	Slots filled, no owner, no rejected options	Convergence must leave “chose / rejected / why”
Whiteboard distortion	Energetic room, no audio, blurry photos, messy stickies	Record + clean board photos before AI; dirty input amplifies ambiguity
AI as decider	“In summary, pick B”—humans nod but cannot restate the trade-off	Close the agent: can you explain the decision to a colleague in three sentences? If not, fake convergence

Minimal action list for the next requirements session:

1. Name which Double Diamond stage you are in before brainstorming.
2. Pick one template as this round’s convergence shape (elevator pitch or journey map is usually enough).
3. Online: multiple candidates → human consensus; offline: whiteboard + audio/photos → AI structure → human decisions.
4. Persist the consensus artifact, then move to Spec / Plan / story slices—do not leave the conclusion only in a chat session.

5. Closing

AI4SE is not “stronger models replace design thinking.” It makes Design Thinking and classic frameworks more executable: rhythm from the Double Diamond, shape from templates, engine from AI brainstorming, decisions from humans.

From rough idea to elevator-pitch consensus, from whiteboard workshop to a retrievable journey-map document—you are not mainly learning to write better prompts. You are learning to converge, in the right space, into the right shape, with a trail that lasts. Tools change; the stack transfers.

References

• Design Council — Framework for Innovation / Double Diamond (CC BY 4.0)
• Geoffrey Moore — Crossing the Chasm (positioning / elevator-pitch structure)
• Jeff Patton — User Story Mapping
• Brainstorming with AI: Why Clarify Beats Prompt
• AI4SE Workshops Are Not Training Classes