The Challenge
A major Australian bank needed a platform to manage Apigee shared flows — the reusable policy bundles underpinning every API proxy across multiple environments (SIT, UAT, Production). The team needed visibility into deployment status, runtime health, error patterns, and blast radius of any change across 625+ proxies.
No off-the-shelf tool existed. Traditional development would require weeks of planning, multiple sprint cycles, and a dedicated engineering resource. Singularity Tech used AI spec-driven development to build it in a single day.
The Approach: AI Spec-Driven Development
Rather than writing code, the developer described capabilities in plain English. Claude generated complete implementations. The developer validated against real production data and gave natural language feedback. Claude refined. The entire product roadmap existed as a conversation.
- 01DescribeThe developer describes the desired capability in plain English — no user stories, no tickets, no pull requests.
- 02GenerateClaude writes the complete implementation: Python modules, API clients, HTML templates, XML parsers.
- 03ValidateThe developer tests against real production data — live APIs, actual environments, 625 real proxies.
- 04RefineNatural language feedback: "the output is hard to read," "add cross-org support." Claude extends within the same session.
What Was Built: Floey
Floey is a comprehensive platform for analysing, monitoring, and managing Apigee shared flows. Eight core Python modules, each 300–780 lines, covering:
- ⚡Deployment Audit & Health DashboardLists all shared flows with revision numbers across every environment. Scores each flow 0–100 across availability, errors, performance, and reliability.
- 🔍Error Intelligence & Blast Radius AnalysisDetects recurring error patterns and spikes. Parallel proxy scanner scans 625 proxies in 2–3 minutes, classifying dependency criticality before any change.
- 📊Performance Profiling & Cross-Environment ComparisonTracks P50/P95/P99 latency. Runs SIT, UAT, and Production health analysis in parallel with side-by-side comparison.
- 📄Multi-Format ReportingAuto-exports JSON, Markdown, HTML dashboards, and colour-coded Excel workbooks — including features never explicitly requested, built by Claude proactively.
"Several of Floey's most valuable features were never explicitly requested. Claude recognised opportunities for improvement and built them as part of its responses. This is the distinguishing characteristic of AI spec-driven development."
Results
| Dimension | Traditional | AI-Driven | Improvement |
|---|---|---|---|
| Full platform build | Weeks to months | ~6 hours | ~95% faster |
| Code authored by human | All of it | Zero lines | 100% AI-generated |
| Documentation | Written after (often skipped) | Generated alongside code | Always complete |
| Incident triage | 30–60 minutes | Single command | ~95% faster |
Zero human-written code doesn't mean zero human involvement. The developer's expertise — knowing the Apigee domain, recognising good output from bad — was what made the project succeed.
AI goes beyond the spec. Bundle caching, cross-org comparison, smart terminal output — all added by Claude without being asked, because it recognised the engineering opportunities.
Documentation is automatic. Because Claude authored both the code and the docs, there is no gap between what the code does and what the documentation describes.
The methodology is repeatable. Every feature followed the same loop: describe → generate → validate → refine. The approach scales to any domain where the human has deep expertise.
Want to see what AI spec-driven development can do for your team?
Singularity Tech delivers production software faster and at a fraction of the cost. The assessment is free.
Talk to our team