Research Synthesis

🔬 Prototype Pilot — Week 2 Synthesis

From setup friction to system friction — what we learned in the second week of the pilot

← View Week 1 Synthesis

TL;DR Summary

Friction Shift
Setup → collaboration + workflow
Prototype Power
Now influencing PM direction & team alignment
Work Pattern
Non-linear, but infra enforces serial coordination
Biggest Gap
Shared infrastructure for parallel prototyping
Key Risk
Cognitive load shifting from product thinking → debugging
Core takeaway: Prototype-first is no longer blocked by tools — it's now constrained by coordination, ownership, and workflow clarity

Quant Snapshot

Effectiveness mixed vs W1
2–5
range  ·  high variance (no single average)
1
0
2
2
3
1
4
2
5
1
Variance driven by environment quality, not just learning curve
AI Comfort bimodal
2–5
range  ·  bimodal distribution
1
0
2
2
3
1
4
1
5
2
Comfort now split: power users accelerating, others still adapting

Interpretation

  • Effectiveness is now environment-dependent, not just learning-dependent
  • Teams with smoother environments are seeing high effectiveness
  • Teams hitting infra/collab friction are scoring significantly lower
Week 1 variance was about onboarding. Week 2 variance is about workflow and infrastructure quality.

Key Themes — Week 2

⚡

Prototypes Are Becoming a Decision-Making Tool

  • Prototypes now influencing PM direction and team alignment
  • Ideas and rationale "played back" from prototypes, not decks
  • Joe's design studio used live prototypes instead of Figma walkthroughs
Step-change from Week 1 — prototypes moved from artifacts to instruments
🔄

Collaboration Friction Is Now the Bottleneck

  • PR conflicts and broken shared prototypes
  • Dependency on others' schedules for merging
  • Andre found remixing others' prototypes broke originals
"Working in GitHub is very intensive… dependent on other people's schedules"
🧩

Parallel Work Breaks the System

  • Multiple people editing shared prototypes causes collisions
  • No isolation — shared slug names caused last-write-wins bugs
  • No safe iteration space for branching and experimentation
Prototype-first assumes parallel exploration — but infra enforces serial coordination
🧪

Iteration Speed Is Real — but Fragile

  • Fast: idea → test → adjust loops when unblocked
  • Fragile: breaks with unstable environments or others' changes
  • Praveen's team used AI to generate bundle compositions as starting points
Speed is proven — but only in protected, stable environments
🧠

Cognitive Load Has Shifted (New Kind of Work)

  • More time on: debugging, merging, infra work
  • Less time on: product thinking, design iteration
  • Model fidelity issues (Claude's output not matching Figma) add overhead
Time is being spent on the wrong layer — infra is taxing product thinking
🧭

Workflow Still Undefined — Now Actively Causing Friction

  • Week 1: "unclear" → Week 2: actively blocking work
  • Questions: how to share, who owns what, when to merge vs branch
  • Teams unclear if contributing one prototype or each team builds separately
Workflow clarity is no longer a nice-to-have — it's a prerequisite

In-the-Moment Feedback

merge conflicts Vercel access namespace collisions Figma MCP debugging overhead sharing externally template requests workflow clarity ownership ambiguity

Key Tensions

Want
Parallel Exploration
vs
Reality
Shared State

Teams want to explore many directions at once, but a shared repo with flat namespaces means edits collide and prototypes break.

Achieved
Speed
vs
Required
Stability

Fast iteration is possible and proven — but fragile environments, broken builds, and dependency on others kill momentum instantly.

Need
Collaboration
vs
Need
Independence

Collaboration across teams is essential for alignment, but creates blocking dependencies. People want to remix work without risking it.

Spending On
Debugging
vs
Want To Do
Product Thinking

More time spent on infra, merging, and fixing Claude's output. Less time on the actual product exploration and design iteration.

Emerging System Insight

Prototype-first is not just a tool shift — it's a system shift

This is the biggest Week 2 realization. The pilot is no longer testing whether people can use prototype tools — it's revealing what system infrastructure is needed to make prototype-first work at scale.

  • Sandboxed environments — so parallel work doesn't break shared state
  • Parallel-safe workflows — branching and remixing without collisions
  • Lightweight sharing — public links for research, internal previews without Vercel auth
  • Infra abstraction — templates, golden path screens, prod-as-starting-point
Current gap: We're running a parallel system on top of a serial infrastructure

Slack Channel Signals (#prototype-first-pilot)

🔧 Friction

  • Prototype slug collisions caused one person's work to overwrite another's (Andre + Praveen)
  • Vercel auth blocks sharing with internal partners not in the pilot
  • Figma MCP authentication issues persisted into Week 2 (Amy)
  • Git commands in Claude Code intermittently broken (Andre)
  • Claude-generated code breaking builds — caught during PR review (Amy's PR)
"You can easily get into a situation where you break the original prototype, even if your code doesn't touch anyone's workspace" — Andre

🚀 Accelerators

  • Praveen's team used AI to generate location/weather-based bundles and parse RF questions
  • Erik shipped namespace isolation fix same-day after Andre flagged the issue
  • GitHub app auto-notification for PRs eliminated a manual step
  • Research sharing workflow created with isolated Vercel deployments
  • Template discussion kicked off — teams requesting golden path screens
"These starting points saved us meaningful time in the early exploration phase" — Praveen

🔮 Emerging Patterns

  • Frank's RFC proposes "holodeck" sandbox for AI-driven prototyping with guardrails
  • Dave's vision post reframed the pilot: testing what tools/workflows we need, not validating one tool
  • Andre wants to pull from prod as starting point instead of Figma
  • Joe's design studio surfaced gaps: no async commenting, no self-paced exploration
  • Teams self-organizing fixes (Andre fixed slug collision, Praveen deployed separately)
"Use whatever is helpful at your disposal… the point is to understand what a prototype-first, AI-native workflow should look like" — Dave

Week 3 Focus

Top Priority
01

Enable Independent Workspaces

The #1 structural unlock. Parallel exploration requires isolated environments.

  • Reduce merge conflicts through namespace isolation
  • Allow safe remixing without breaking originals
  • Build on Andre's namespace fix as a foundation
02

Improve Sharing Layer

Broader access to prototypes — for internal partners, research participants, and stakeholders.

  • Public links without Vercel authentication
  • Isolated research deployments
  • Reduce dependency on deployment bottlenecks
03

Define Collaboration Model

Clarify the rules of engagement so teams can move fast without stepping on each other.

  • Who owns what prototype or workspace
  • When to branch vs merge vs remix
  • How to work in parallel across teams
04

Reduce Debugging Overhead

Protect prototyping time by making the environment more predictable.

  • Guardrails for AI-generated code quality
  • Stable base environments and templates
  • Golden path screens as starting points
05

Re-Center on Product Thinking

Reduce the infra tax so people spend time on what matters — exploring product ideas.

  • Templates and pre-built views (customer golden path)
  • Pull from prod as starting point, not just Figma
  • Protect dedicated prototyping time from infra overhead