Anton Yefimenko
I'm Delivery Director at QATestLab, where I run about five QA programmes with about 140 people. I became Program Director in November 2022, grew delivery from about 35 to about 80 people, and took the director role in June 2025. Before that I ran programme delivery for a large fintech client at SupportYourApp (2019 to 2020) and managed projects at ESL FACEIT.
Things I think about
- AI agents that check their own work
- When test automation pays for itself
- Scaling a delivery team without losing quality
- Evidence over demos
- What QA buyers actually pay for
Selected work
Working builds for real QA and sales work. Numbers appear only where they've been measured.
Test-case automation in n8n
Four agents in n8n take on a recurring weekly task across 400+ test cases. A person checks the output before it's used.
About 10 hours a week down to about 1, with about 30% fewer errors.
Self-verifying QA agent swarm
Three research agents probe a live e-commerce system without pause and must cite logged calls as evidence. A rules-based judge drops unsupported claims, and a fourth agent replays every finding before a person sees it.
On a test shop with 5 seeded defects and 6 decoys, the verifier confirmed 3 of 3 defects and none of 4 decoys in three runs.
QA cost calculator
Compares the full monthly cost of an in-house QA specialist with an outsourced team. Prospects enter their own salary, taxes, overhead, idle time and leave. The maths runs in the browser and nothing goes to a server. A free tool for QATestLab sales.
Try it (opens in a new tab)Automation scope estimator
Ten questions in, a pilot plan out: recommended scope, effort in hours by phase, a calendar range, a five-point readiness check and current manual regression hours. Every coefficient is shown on the result page. Built for QATestLab's test automation service.
Try it (opens in a new tab)Sales signals digest
Scans app store reviews, job openings and news for client and prospect companies, and turns fresh signals into reasons for QATestLab sales to call. Every row needs a source link and a signal date. A self-check script holds back anything unsourced.
In one weekly run, 63 companies went in and 22 came out with fresh signals.
Signal Academy
A training platform for SDR and lead-gen teams with no testing background. Learners practise presale calls with AI prospects whose objections adapt to the conversation, then get scored on four criteria. A skills page tracks evidence across seven skills, and a mentor makes every pass or retry call.
Speaking & advisory
I'm open to talks, panels and advisory conversations on AI in software testing and QA delivery at scale. Topics I can go deep on: where AI agents earn a place in a test process, how to check what an agent reports, and how to run several QA programmes without losing quality. Send me a message on LinkedIn.