Quality assurance · Production

ASOS introduces Test-Driven Vibe Development (TDVD) to deliver MVP 7–10× faster than estimate

The problem

Vibe coding with AI agents produced code without sufficient rigour, with hallucinations, hidden defects, and security risks. LLMs' attention diffusion in large contexts caused accuracy to decline as prompt complexity grew, making naive AI-assisted development unreliable for enterprise product builds.

First attempt

Initial vibe coding attempts matched widely documented concerns about quality and rigour. During the TDVD trial, early over-sized feature prompts caused LLM context loss, requiring the team to tear down and rebuild the solution twice.

Workflow diagram · grounded in source
1
Plan: define intent test-first
trigger
“Plan — defining the intent in a test-first manner”
2
Generate acceptance tests via AI
ai_action
“Clear testable requirements as code are generated before functional implementation”
3
AI generates functional code
ai_action
“Code generation is aligned with testable requirements”
4
Continuous validation against tests
validation
“AI-generated implementations are continuously validated against those requirements”
5
Refine prompts and user stories
feedback_loop
“We adapted by revising our PRD by breaking the capability features down into much smaller user stories, and rebuilding our implementation step guides so that each cycle focused on no more than two features”
6
Harden with deeper tests
validation
“Harden — final layer of deepening functional and non-functional test depth”
Reported outcome

The team delivered the MVP plus additional capabilities in 4 weeks, which was 7–10 times faster than the original estimate of 4–6 months, with 42 hours of total active development time.

Reported metrics
Development speed vs original estimate7–10 times faster
MVP delivery time4 weeks
Total active development time42 hours or 11 workdays
original MVP estimate4–6 months
Show all 6 reported metrics
development speed vs original estimate7–10 times faster
MVP delivery time4 weeks
total active development time42 hours or 11 workdays
original MVP estimate4–6 months
MVP rebuild percentageapproximately 80%
mob session hours32 hours
Reported stack
Azure FunctionsReactAI agent
Source
https://medium.com/asos-techblog/introducing-test-driven-vibe-development-0effe6430691
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The team delivered the MVP plus additional capabilities in 4 weeks, which was 7–10 times faster than the original estimate of 4–6 months, with 42 hours of total active development time.

What tools did this team use?

Azure Functions, React, AI agent.

What results were reported?

Development speed vs original estimate: 7–10 times faster; MVP delivery time: 4 weeks; Total active development time: 42 hours or 11 workdays; original MVP estimate: 4–6 months (source-reported, not independently verified).

What failed first in this deployment?

Initial vibe coding attempts matched widely documented concerns about quality and rigour.

How is this quality assurance AI workflow structured?

Plan: define intent test-first → Generate acceptance tests via AI → AI generates functional code → Continuous validation against tests → Refine prompts and user stories → Harden with deeper tests.

Related quality assurance cases