A2D2 Framework - Double diamond in the age of AI
AI-Augmented Double Diamond framework for the age of AI
The Double Diamond framework remains a trusted foundation, but it’s time to refresh it for the AI era. Its core principles still hold true. Yet, by integrating AI natively, while keeping human judgment central for decisions and context, we can work smarter, faster, and more effectively.
This is my take on the how AI-augmented Double Diamond framework might look like in its final form.
A2D2 Framework
A2D2 keeps the classic “Discover → Define → Develop → Deliver” flow, but straps an AI engine to the front of each diamond and inserts clear human “gate” moments so the team always steers—not the model.
How it works
Data Pulse (the pre-diamond runway)
Instead of kicking off with a brainstorm, you start by funneling every scrap of real customer voice like support tickets, chat transcripts, survey comments into an AI pipeline.
The model clusters similar complaints, tags urgency and sentiment, and spits out a ranked list of “problem clusters,” each with sample quotes attached.
Human Gate A: A product or design lead skims the clusters, merges or renames a few, and crowns one as the focus for this cycle.
Discover (first diverge)
AI now deep-dives on that chosen cluster: pulling related analytics, competitor reviews, Reddit chatter, anything relevant.
It drafts a discovery brief that explains who’s hurting, why, and where the evidence comes from.
Human Gate B: A small trio (PM, designer, researcher) sanity-checks the brief. Adds context the model couldn’t know, like business constraints, technical quirks… And signs off.
Define (first converge)
The same model reframes the brief into crisp “How Might We…?” statements, short user stories, and success metrics.
Human Gate C: Stakeholders pick the wording that actually resonates and lock it as the official Problem Statement. From here on, scope creep is easier to spot because anything outside that statement is “out of bounds.”
Develop (second diverge)
Generative AI floods the room with possibilities: sketches, UX flows, even quick image mocks. Quantity first, quality later.
Human Gate D: The design team runs a quick dot-vote or workshop, curates the pile, and chooses at least one concept (ideally two or three) to detail.
Deliver (second converge)
AI expands the chosen concept(s) into a living mini-spec: edge-case notes, copy draft, acceptance criteria, hand-off checklist.
Human Gate E: Engineering, QA, and PM review feasibility, tweak as needed, and drop the spec into the roadmap. Build begins.
Why teams will like it
Evidence over ego: You start with thousands of real user voices, not the loudest opinion in the room.
Right-sized AI: Models do the heavy lifting, while humans keep every strategic decision.
Built-in traceability: Each AI output carries links to its source data, so claims can be audited in seconds.
Familiar shape: Because it’s still a Double Diamond, veterans grasp the flow instantly; newcomers see exactly where AI plugs in.
Let me know you think about it.
btw. I will try make a simple tool that would be built around A2D2 framework. Let me know if you would be interested trying it out once is done.