From SDET to Agentic QA Engineer
2026-07-01
Five years ago I was writing Cucumber features at Newfold Digital.
Today I'm building AI agents that write test suites for me.
Same job title (QA Engineer). Completely different job.
The shift happened gradually:
- 2021: Protractor/JS → Cucumber (over WebdriverIO)
- 2022: Maintained and refactored the Cucumber suite — regression cut from 3.5–6 h to ~1:45–2 h (~60%)
- 2024: Re-platformed to Robot Framework + Python · agentic workflows — AI writes tests, I review
- 2026: Built the Agentic QA Boilerplate (KATA + Playwright) — open-source, with a create-agentic-qa scaffolder, 20+ agent skills, a full Xray CLI, and 7 MCP integrations
- 2026: Launched career portfolio at nelthor.com.ar — built via agentic dev workflows
The Newfold Years (2021–2026)
I joined Newfold Digital as an SDET working across consumer-facing products — networksolutions.com, bluehost.com — serving millions of users. My first task was taking over a 600+ test suite and making it maintainable.
We started by migrating it from Protractor/JS to Cucumber over WebdriverIO, then I refactored that suite on my own initiative — cutting regression from 3.5–6 h to ~1:45–2 h (~60%) while trimming ~600–630 → ~430–440 cases. Around 2024 we re-platformed to Robot Framework + Python, shifting UI coverage to faster API/DB (headless) tests — runs under 1:30 h with equal coverage and higher precision.
The key lesson: framework architecture matters more than tool choice. A well-structured test suite survives team changes, product pivots, and technology refreshes. A poorly structured one becomes a liability.
The Agentic Shift
In 2024 I started experimenting with AI-assisted test generation. What began as "let me see if ChatGPT can write a Playwright test" turned into a full agentic workflow:
- AI proposes test scenarios from natural language intent
- I review for correctness, coverage, and edge cases
- AI generates the Playwright/Cucumber code
- I approve and merge
The bottleneck shifted from writing code to designing intent. That's the agentic QA difference.
What I Built
The KATA (Component Action Test Architecture) framework emerged from this work — a three-layer pattern:
| Layer | Purpose | Who owns it |
|---|---|---|
| Component | What you test | Dev |
| Action | How you interact | Framework |
| Test | What you assert | AI + Human |
AI fits naturally at Layer 3: it generates assertions from natural language intent while the abstraction layers keep the framework stable.
That framework grew into the Agentic QA Boilerplate — an open-source starter that turns the whole QA lifecycle into agent skills. Instead of isolated test snippets, it orchestrates the loop: shift-left testing (Stage 0), in-sprint manual QA (Stages 1–3), test documentation and ROI prioritization (Stage 4), KATA automation (Stage 5), and regression with GO/CAUTION/NO-GO sign-off (Stage 6). It ships with a create-agentic-qa scaffolder, 20+ skills, a 13-command Xray CLI for Jira/Xray traceability, and 7 MCP integrations.
The lesson stayed the same: design your framework for AI collaboration, not just human maintenance.
Looking Forward
Agentic QA is not about replacing engineers. It's about shifting the work from writing code to directing quality. The AI handles the mechanics; the engineer handles the intent.
That's what Agentic QA Engineer means to me.