In a Perforce industry survey published in 2025, 75% of respondents identified AI-driven testing as a pivotal component of their strategy. Only 16% had actually adopted it. That 59-point chasm between intention and execution is the defining feature of quality engineering in 2026 — and it is closing fast, in one direction only. The World Quality Report 2025-26, published by Capgemini and Sogeti, found that 89% of organisations are now piloting or deploying generative-AI-augmented QE workflows, with 37% already in production. Manual QA, as a standalone profession, is being squeezed between a technology that works and an industry that has finally decided to use it.
The thesis is not that manual testing will vanish overnight, but that its economic rationale has collapsed. When AI tools can generate exploratory scenarios, self-heal broken locators, and adapt to UI changes autonomously, the case for staffing large manual QA teams evaporates. Companies clinging to headcount-heavy manual processes are not exercising caution — they are burning capital.
The economics of extinction
The numbers are brutal. The Capgemini report found that the rate of non-adopters of generative AI in quality engineering dropped from 31% in 2023 to just 11% in 2025. Among adopters, test-cycle time reductions of 30-40% were commonplace. Tricentis’s 2025 State of Testing survey confirmed that three-quarters of teams using traditional code-based automation frameworks had already layered AI tools on top to assist with test writing and maintenance.
Yet the adoption gap persists because the barriers are organisational, not technical. The top obstacles cited in the World Quality Report were hallucination and reliability concerns (60%), data privacy risks (67%), and integration complexity (64%). These are governance problems, not engineering problems — and governance problems dissolve the moment a competitor ships faster because it solved them first.
The AI competence threshold
Two breakthroughs have shifted the calculus. First, large language models can now generate meaningful exploratory test scenarios — a capability that was the exclusive preserve of experienced human testers as recently as 2023. Google DeepMind researchers demonstrated in 2025 that LLM-based test generation achieved 82% branch coverage on previously untested Java codebases, competitive with experienced human testers given the same time constraints. Applitools demonstrated in late 2024 that its Visual AI engine detected 31% more visual regressions than senior manual testers across a benchmark of 2,000 screen comparisons, producing fewer false positives.
Second, self-healing test frameworks crossed the reliability threshold. Katalon’s 2025 benchmark showed that AI-maintained locators survived 89% of UI changes without human intervention, up from 54% in 2023. The maintenance burden that made automation expensive — the very argument manual QA advocates wielded for decades — is dissolving. Deloitte projects that 25% of businesses investing in generative AI will deploy autonomous testing agents in 2026, rising to 50% in 2027.
The workforce reckoning
The human consequences are stark and uneven. Demand for manual test execution roles has cratered while demand for test automation architects and AI/ML quality engineers has surged. The pattern repeats across the industry: Spotify’s quality engineering team restructured in mid-2025, eliminating dedicated manual QA positions entirely and embedding quality ownership within cross-functional squads equipped with AI testing toolchains. Atlassian, Shopify, and Stripe have disclosed similar reorganisations.
The common thread is not mass layoffs but role transformation. Former manual testers who upskilled into automation or AI-assisted testing retained their positions. Those who did not found their roles consolidated or eliminated. As Jason Arbon, CEO of test.ai and former Google engineering director, has argued publicly: “The biggest barrier to AI testing isn’t the AI — it’s the ten-year-old test suite nobody wants to touch.”
The uncomfortable conclusion
Manual QA is not dying because it was bad. For decades it was the only reliable method for validating software behaviour, and the discipline produced rigorous thinkers who understood systems deeply. But economics is merciless. The 77.7% of QA teams that have adopted AI-first quality approaches, according to the 2026 QA Trends Report, are not early adopters anymore — they are the new mainstream.
The remaining holdouts face a compounding disadvantage. Every quarter that a competitor runs AI-assisted regression suites while a rival’s manual testers click through the same screens, the gap in release velocity and defect economics widens. The companies that thrive will be those that redeploy human intelligence where it genuinely outperforms machines: understanding user intent, questioning ambiguous requirements, designing quality into architecture rather than inspecting it after the fact, and — critically — supervising the AI tools that are doing the inspecting. The rest will discover that nostalgia for manual processes is an expensive indulgence in a market that stopped subsidising it the moment the machines learned to explore.
References
- Perforce Software, State of Continuous Testing 2025, perforce.com
- Capgemini and Sogeti, World Quality Report 2025-26, capgemini.com
- Tricentis, State of Software Testing 2025, tricentis.com
- ThinkSys, QA Trends Report 2026: Market Growth, AI-Driven Testing, thinksys.com
- Rainforest QA, AI in Software Testing: State of Test Automation Report 2025, rainforestqa.com
Discussion