TL;DR Summary
Effectiveness
Avg 4.0/5 — highest of the pilot, up from 2.5 in Week 1
AI Comfort
4/4/4 — fully converged, learning curve is done
Biggest Win
Researcher built working conversational prototype, ran live user testing
Remaining Friction
Git/merge workflow still too heavy for lightweight iteration
What's Next
Scaling conversations started — expanding access beyond pilot
Core takeaway: Prototype-first is generating real product learning and changing who can make things. The playbook question is no longer whether this works, but what infrastructure it needs to scale.
Pilot Evolution
How the bottleneck shifted each week
1
Week 1
Setup & tooling
2
Week 2
Collaboration & infra
3
Week 3
Handoff & process
4
Week 4
Scale & playbook
Quant Snapshot
Effectiveness ↑ 4.0 avg
4.0
out of 5 · range: 3–5
First time no one scored below 3. The floor has risen.
Trend
2.5
W1
~3.3
W2
3.0
W3
4.0
W4
AI Comfort converged at 4
4.0
out of 5 · range: 4–4 (all identical)
Second consecutive week at 4.0 — AI tools no longer a learning barrier
Trend
3.2
W1
~3.4
W2
4.0
W3
4.0
W4
Interpretation
- Effectiveness jumped to its highest point — 4.0 avg vs 2.5 in Week 1
- The floor rose: no scores below 3, first time in the pilot
- AI comfort is a solved problem — steady at 4/5 for two weeks running
- The 5/5 came from a researcher who built a working conversational prototype with an engineer — the clearest proof of prototype-first value so far
Effectiveness is no longer constrained by tools, comfort, or setup — it's now a function of how well the workflow supports the team's specific use case.
Key Themes — Week 4
Prototypes Generating Real Product Learning
- Live conversational prototype revealed nuance static message testing wouldn't have
- Prototypes exposed fundamental assumption flaws, expanding team scope
- Synthesizing months of research and context into divergent explorations quickly
This is the thesis confirmed: prototype-first accelerates learning by making ideas real sooner
Researcher as Maker
- A researcher made fast stimulus changes directly instead of waiting on design/eng
- Partnered with engineer to calibrate simulated pro behavior for testing
- "Researcher-accessible prototyping tools are really powerful for tightening stimuli"
Prototype-first doesn't just speed up the process — it changes who can participate in making
Git/Merge Workflow Is Still the #1 Friction
- All three respondents mentioned merge/branching pain
- "The whole Git process still feels finicky and manual — errors, branch confusion"
- Merging small changes created more back-and-forth than expected
The path from "slightly revised prototype" to research deployment needs to be much lighter weight
Remix & Versioning Workflow Needed
- Want to branch off ideas from main prototype without breaking it
- Explore variations without bloating navigation or risking the main file
- Parallels the "sandboxed environments" ask from Week 2, now more specific
The system needs to support divergent exploration as a first-class workflow, not a Git workaround
Figma MCP Hit a Plateau
- Still slow, requires maintaining focus on Claude's output window
- One participant installed a Chrome extension that was faster with more manual work
- Workarounds emerging naturally — teams adapting around the limitation
Figma MCP needs performance investment or the ecosystem will route around it
Scaling Conversations Have Started
- Design Systems team discussing post-pilot ownership and access expansion
- Vercel enterprise and Google Cloud Run alternatives being evaluated
- New engineer (Michael) getting access to contribute to the playground
The pilot is already outgrowing its container — a good problem to have
In-the-Moment Feedback
What's Working
- Synthesizing months of work into divergent prototype explorations
- Prototypes pointing out fundamental assumption flaws
- Researchers making fast stimulus changes directly
- Live conversation prototypes revealing more nuance than static testing
- Starting to feel like a better way to begin the design process
What's Breaking
- Git pull/merge process still finicky, manual, and error-prone
- Figma MCP slow, requires focus lock on Claude's window
- Merging small changes into main takes too much back-and-forth
- No remix/versioning workflow for exploring variations safely
- No project documentation layer outside the prototype
High-Value Signals
- Lighter path from revised prototype to research-ready deployment
- Remix workflow: branch off ideas without breaking main prototype
- Claude Cowork / enterprise desktop app access
- More automation in publishing to Vercel
- Commenting and easier sharing (still unresolved from Week 3)
git/merge friction
remix workflow
Figma MCP speed
research deployment
Claude Cowork
publish automation
async commenting
divergent exploration
Key Tensions
Actual
Git Overhead
vs
Needed
Lightweight Publish
Every deployment requires a PR, merge, and branch management. Teams want a one-click path from tweak to shareable prototype.
Proven
Learning Speed
vs
Question
Skipped Diligence
Prototyping fast reveals real product gaps. But one team noted they "would have done more due diligence before asking design to make a hi-fi prototype." When does speed skip necessary thinking?
Emerging
Broader Roles
vs
Familiar
Normal Week
One respondent did coding and PM-like scoping outside their role. Another said "this week felt like a pretty normal design week." The method lands differently depending on the role.
Pilot
Contained Scope
vs
Reality
Ready to Scale
Design Systems is already discussing post-pilot ownership, access expansion, and Vercel enterprise. The pilot is outgrowing its container before it's officially over.
Final Week Insight
The pilot proved the thesis — now the question is infrastructure for scale
Four weeks ago, the question was "can non-engineers build interactive prototypes with AI tools?" That's been answered: yes, and they're generating product insights that wouldn't have emerged otherwise. A researcher built a working conversational prototype. Teams pivoted based on what prototypes revealed. Months of planning were synthesized into divergent explorations in days.
- The learning curve is solved — AI comfort at 4/5 for two straight weeks
- The value is proven — prototypes driving real product decisions and user signal
- The friction is structural — Git workflow, deployment overhead, Figma MCP speed
- The next phase is organizational — who owns this, how does it scale, what infra is needed
The pilot answered "does prototype-first work?" The playbook needs to answer "how do we make it work for everyone?"
Slack Channel Signals
🚀 Product Work in Motion
- DaRe team used prototypes to explore 7 response quality archetypes as live conversations
- Joe shipped expanded prototype states (expanded sheet, clear action, saved replies)
- Praveen built individual agents for user research, data research, market research, and prompt building
- Dave wants walkthrough of Praveen's v1 tool — interest in reusable patterns
"Even being able to type into a focused state on mobile and have it respond was magic" — Dave (previous week, still resonating)
🔧 Platform & Infra
- Erik merged sizable cleanup removing unused/non-necessary code
- Design Systems discussing post-pilot: expanding access, Vercel enterprise evaluation
- Michael Movsesov added as admin to prototyping-playground repo
- Brian exploring Vercel enterprise; Erik open to Google Cloud Run but flagging tradeoffs
- Erik: "Following phase 1 of scaling AI-driven product development"
"I'd like to get Michael up to speed next week to contribute. Short term would be optimizing the experience and expanding access beyond the scoped pilot." — Erik
🔮 Signals for the Playbook
- Research-ready deployment is a key workflow that needs a lighter path
- Spinning up and tearing down research branches is a core capability to preserve
- The playground is being discussed as a lasting tool, not a pilot artifact
- Other teams (outside pilot) already requesting prototyping-playground access
- Chris (returning from OOO): "really fun to see all the progress" — external validation
"One of the larger challenges will be spinning up and tearing down research branches. We've enabled this in Vercel but it might be an additional learning curve for other platforms." — Erik
Playbook Priorities (Post-Pilot)
Top Priority
01
Simplify the Publish & Deploy Path
The heaviest remaining friction. Git PR workflow is too much overhead for lightweight iteration and research sharing.
- One-click path from prototype tweak to shareable URL
- Lighter branching for research variants
- More automation in Vercel publishing
02
Build the Remix/Versioning Workflow
Teams need to branch off ideas from a main prototype without breaking it or bloating the project.
- First-class support for divergent exploration
- Safe variation testing without merge risk
- Evolves the "sandboxed environments" ask from Week 2
03
Define the Handoff Model
Carried over from Week 3 — still unresolved. What does engineering receive when prototypes lead?
- Can prototype code serve as a starting point for eng?
- What minimum Figma (if any) is needed?
- Document the recommended path in the playbook
04
Invest in Figma MCP Performance
The Figma-to-code pipeline works but is too slow. Teams are already routing around it with Chrome extensions.
- Speed up the MCP connection and frame reading
- Remove the focus-lock requirement on Claude's window
- Or formalize the faster workaround paths
05
Plan the Scale Path
The pilot is already outgrowing its scope. Prepare for broader EPAD rollout with clear ownership and infra.
- Resolve Vercel enterprise vs alternative hosting
- Define ownership model for the playground post-pilot
- Onboarding path for teams not in the pilot
- Async feedback and commenting layer