The New Skill-Currency:
TDD, E2e & AI Power Tools

Murat K. Ozcan

Senior QA Architect

Ambassador

Combinatorial Testing Comittee

@MuratKeremOzcan

SEON: the command center for

fraud prevention and compliance

 

Contents

  • AI for test generation

  • TDD-first AI flow

  • AI Customization & Orchestration

  • The new skill currency

Does AI Actually Boost Productivity?

AI for small-scope tests

 

Great for unit tests / single-file context

code sample

AI for mid-scope tests

 

Decent for integration tests

with a few files, even in multiple repos

code sample

AI for large-scope tests

 

Larger context -> worse experience

many files

network(?)

system knowledge

// Ideal e2e test:

// spy on network
cy.intercept('..')

// do things
cy.get('..').click()

// cross verify network vs UI
cy.get('..').should('..')

code sample

Contents

  • AI for test generation

  • TDD-first AI flow

  • AI Customization & Orchestration

  • The new skill currency

TDD-first AI Workflow

hold the reins

Write the test

Ask AI to implement

Run the test

Refine || Reject

Repeat

🔒 TDD as guardrail:

Every AI generated change

must pass these tests before proceeding

TDD-first AI Workflow

 

Write the test

code sample

TDD-first AI Workflow

 

Ask AI to implement

code sample

TDD-first AI Workflow

hold the reins

Run, Refine || Reject, Repeat

Contents

  • AI for test generation

  • TDD-first AI flow

  • AI Customization & Orchestration

  • The new skill currency

AI customization & orchestration

 

Use Rules

murat's rules

AI customization & orchestration

 

Use memories

(let it go through everything...)

note: Claude Code combines rules & memories:

/init (CLAUDE.md) + # memory

gitingest.com

npx bmad-method flatten

AI customization & orchestration

 

Use MCPs, not too many

AI customization & orchestration

 

Use planning mode/todo list, or full context-engineering

BMad Method

agentic planning: analyst, pm, arch. -> PRD, arch.spec

context-engineed development: SM, dev, Qa -> epic, story, task

eliminates both planning inconsistency and context loss

AI vs AI review loop

 

Get a 2nd opinion: Model A writes code, Model B reviews

Windsurf PR reviews, ChatGPT Codex, Copilot for PRs, CodeRabbit etc.

/code-review

Contents

  • AI for test generation

  • TDD-first AI flow

  • AI Customization & Orchestration

  • The new skill currency

The new skill currency

 

Precision framing - nail down the exact problem boundaries, provide the context, so AI zeroes in on what matters

Quality radar - spotting good enough code & refining it

Debugging & troubleshooting - ambiguous bugs, flakey tests

Creative thinking - finding unconventional solutions that scale

Persuasion & collaboration - explaining ideas and rallying others

AI wishlist

 

MCPs to integrate with Test framework runner/browsers

for richer context & better AI feedback on test failure
(network, console, env vars, DOM, UI...)

Better and more AI vs AI interop:

repo-aware Codex + ChatGPT

PR reviews with local context, multi-model code reviews.

Better repo awareness:
ex: how Claude Code initializes (CLAUDE.md file)

 

Higher MCP tool limits

What have we learned

With AI, End-to-End testing is King

AI nails unit & small integration tests, and E2E suite holds the reins

TDD-First/Fast-follow-testing Mindset Puts You in Charge
Define your acceptance tests up front, let AI code to them, run your checks, then refine or reject.

AI Power Tools & Soft Skills Are Your New Currency

Layer in rules, memories, planning, BMAD/context-eng & AI-vs-AI reviews

Master precision framing, quality-radar, debugging & creative collaboration

Made with Slides.com