Overview
Build Agent enables users to create enterprise applications using natural language. While the AI experience promised instant app creation, the underlying platform required users to navigate instance provisioning, version compatibility, workspace creation, role assignment, and IDE transitions before they could even begin.
This project focused on redesigning that activation journey, reducing friction, maintaining user confidence during infrastructure delays, and creating a scalable adoption experience for both first-time and returning users.
The full adoption journey: from first prompt on the Developer Home, through guided instance provisioning, to a ready-to-build environment.
My role
I led the end-to-end UX strategy for Build Agent adoption across the Developer Portal and the Studio / IDE handoff experience. My work focused on turning fragmented platform states into a clear activation journey that helped users preserve intent, understand progress, and reach app creation with confidence.
What I led
- Activation journey strategy
- First-time and return-user flows
- System-state mapping
- Interaction design and prototypes
- Design reviews with product and engineering
Collaboration
- Partnered with PM and TPM
- Aligned with engineering on provisioning constraints
- Reviewed feasibility of auto-configuration
- Worked through edge cases and delivery phasing
Artifacts
- Journey maps
- Modal interaction models
- Hi-fi prototypes
- Design handoff specs
The problem
The promise of AI was immediate app creation, but the platform required users to complete several infrastructure-dependent steps before they experienced any value.
Between writing a first prompt and seeing a generated application, users encountered version selection, role configuration, workspace setup, instance provisioning, waiting states, and IDE transitions. Each interruption created uncertainty and introduced another opportunity for abandonment.
Version confusion
Users could select unsupported platform versions, which prevented Build Agent from launching and made availability feel inconsistent.
Silent provisioning
Instance setup and wake-up could take several minutes with limited feedback, making users feel the system had stalled.
Fragmented handoff
Users moved between Developer Portal, PDI, Studio, and IDE without a clear explanation of what was happening next.
Research & discovery
We combined self-trials, internal feedback, community signals, and engineering reviews to understand where adoption was breaking. The biggest insight was that the issue was not only usability, it was expectation mismatch.
Self trials
We attempted to build with Build Agent ourselves to identify real setup friction and unclear handoffs.
Community feedback
Users reported confusion about availability, plugin / trial access, prompt limits, and where Build Agent actually lived.
Stakeholder reviews
PM and TPM helped separate design-owned opportunities from platform or backend constraints.
Benchmarking
We compared expectations from AI builders like Lovable-style experiences against enterprise platform realities.
Key insights
Users expected AI immediacy
The mental model was “I type a prompt, AI builds,” but the experience required setup before value.
Waiting felt like failure
When the system was quiet during provisioning, users could not tell whether progress was happening.
Prompt quality shaped success
Users needed help moving from short “vibe” prompts to structured enterprise-grade prompts.
Return journeys lacked continuity
Previous apps, prompt usage, and next best actions were not visible from the Developer Portal.
Journey redesign
The strategy was to move from a blocker-heavy setup path to a guided activation journey. The system should preserve the user’s intent, configure what it can automatically, explain what is happening, and transfer users to Studio only when they are ready to build.
Before: high-friction activation
After: guided activation
AI activation states
A core part of the work was designing for infrastructure uncertainty. Instead of treating each backend state as a separate error, we created a system of reusable activation states that could scale across entry points.
Provisioning
Role check, version selection, instance setup, workspace creation, and redirect progress.
No capacity
Waitlist and no-compatible-instance states that give users a next step instead of a dead end.
Recovery
Waking-up instance, setup failure, wake-up failure, dismissed modal, and persistent progress bar.
Key design decisions
Preserve user intent across sign-up
Problem: Users typed a prompt, signed in, and had to restart or click Create again.
Decision: Keep the prompt visible and continue the flow after sign-up.
Assumption: Reduces restart frustration and increases first-build completion.
Auto-configure where possible
Problem: Users had to understand admin roles, instance versions, and workspace setup.
Decision: Let the system configure defaults and explain them during progress.
Assumption: Reduces configuration errors and support dependency.
Turn waiting into education
Problem: Provisioning delays made users feel blocked.
Decision: Use waiting states to show progress, explain platform setup, and teach better prompting.
Assumption: Reduces drop-off during wait and improves prompt quality.
Turn dead ends into next steps
Problem: No instance availability or unsupported versions created abandonment points.
Decision: Offer waitlist, continue-without-BA, request-later, and persistent status options.
Assumption: Keeps users in the adoption loop even when infrastructure is not ready.
Make returning users feel continuity
Problem: Users could not easily find previous builds, prompt usage, or next actions.
Decision: Introduce Build Agent app cards, prompt counters, app status, and continue-building actions.
Assumption: Improves repeat usage and supports upgrade / buy-in moments.
Guide enterprise-grade prompting
Problem: Users expected short prompts to create complex enterprise apps.
Decision: Add advanced prompt guidance with structured inputs and examples.
Assumption: Increases successful outputs and reduces wasted prompts.
Final experience
The final experience worked as a guided activation platform: helping users discover Build Agent, start with a prompt, understand provisioning, recover from unavailable states, and continue building from previous work.
1. AI-first homepage
Build Agent becomes the primary creation entry point with clearer value messaging, suggested prompts, and reduced hero clutter.
2. Contextual sign-up
Sign-up is framed around creating with Build Agent and preserves the user’s prompt instead of restarting the flow.
3. Guided setup
Provisioning explains role, version, instance setup, workspace creation, and Studio handoff in one coherent progress model.
4. Recovery states
Waitlists, unavailable instances, wake-up states, and failures provide clear next actions and status visibility.
5. Returning dashboard
Users can see previous apps, prompt usage, statuses, expiry, and continue-building actions from the Developer Portal.
6. Prompt guidance & Studio handoff
Advanced prompt structure teaches users to describe purpose, data, workflows, UI requirements, and business context, then continues into Studio / IDE depending on instance version.
Reflections
Designing AI products taught me that trust is built before intelligence. Users rarely abandon because AI fails; they abandon when they do not understand what the system is doing.
By designing around infrastructure uncertainty instead of hiding it, we transformed unavoidable technical delays into opportunities for guidance, education, and confidence building.
This project also reinforced the importance of aligning product, platform, and engineering teams around shared user outcomes instead of isolated technical milestones.