After more than 25 years working in software quality across industries including telecommunications, energy, financial services and consumer electronics, I’ve seen plenty of ‘next big things’ come and go.
Right now, that spotlight is firmly on AI. And yes – AI is everywhere. Organisations across every sector are looking at how to use it within their innovation strategies, particularly in software development and testing, where pressure to move faster while maintaining quality continues to climb.
But AI adoption isn’t as simple as switching on a tool.
In reality, what I see time and time again is this: when organisations adopt AI without structure, they create noise. Teams pick different tools, individuals experiment in silos and what you end up with is a landscape of disjointed efforts. Small pockets of value for individuals at best, but no coherence, no consistency and nothing that can scale across the business. The result is more distraction than transformation, and often the build-up of technical debt rather than its reduction.
At Resillion, as a Total Quality company focused on helping organisations bring secure digital products to market faster and at the highest possible quality, we’ve taken a different approach. My team and I have spent the last few years applying AI across quality engineering, cyber security and conformance in a way that remains firmly rooted in engineering discipline. The goal has never been to ‘use AI everywhere,’ but to apply it where it genuinely adds value, while ensuring humans remain in control.
When AI becomes part of the problem
The uncomfortable truth is that AI can easily become part of the problem instead of the solution.
I’ve seen organisations rush to deploy it to fix testing bottlenecks, only to create new risks – false confidence from unvalidated outputs, focus on low-value automation while critical flows go uncovered, and teams spending more time managing tools than improving quality.
QA and QE leaders face relentless pressure to move faster, increase coverage, and reduce risk amid skills shortages and growing complexity. AI looks like the silver bullet, but without structure, governance and human validation, it only adds more complexity and erodes trust.
The real question isn’t whether to adopt AI in software development and testing processes. It’s how to use it without letting it become another source of risk.
Turning chaos into control
We firmly believe that sustainable AI adoption must be treated as a form of software engineering, underpinned by the same core disciplines that govern any successful technical transformation.
Since AI only creates value when it’s focused, rather than deploying tools reactively, we align on specific use cases where AI can remove friction or reduce risk. We apply value mapping and feasibility assessments to sequence initiatives that deliver early wins while staying aligned to long-term quality goals.
Solid technical foundations
While the open availability of AI tools is well suited to community development, a free-for-all approach to AI tooling rapidly becomes unmanageable. We converge on defined AI stacks, data infrastructures, and lifecycle tools that can be governed, secured, tested, and scaled. Our partnerships with hyperscalers and model providers sit within a controlled architectural framework, not as experiments, but as production-grade platforms.
The AI landscape shifts constantly, and it pays to be Agile. Rapid prototyping, MVP development, pilots, and controlled scaling allow us to learn quickly while remaining disciplined. Fast learning is valuable – fast failure without structure is not.
No AI initiative proceeds without measurable outcomes. Business metrics and technical KPIs guide investment decisions and ensure adoption remains tied to meaningful improvement – not novelty.
Governance and risk management
A final word on that increasingly hot-topic for AI – responsible AI development. Every AI innovation is assessed ethically, operationally, and from a security and compliance perspective. Traceability, explainability and auditability remain non-negotiable in environments where quality and trust define brand value.
AI in action: Real use cases across QE
By applying those principles, we’ve focused our work on use cases that consistently demonstrate impact across client environments.
Validating requirements – Stopping defects before they exist
We use natural language processing to analyse user stories, regulatory texts, and technical specifications for clarity, completeness, and contradiction before development even begins. This early quality gate stops ambiguous or incomplete requirements from feeding downstream defects.
In one financial services client programme, this approach reduced live production defects by up to 75%, while also cutting remediation costs across both testing and development cycles.
Human guardrails remain essential. AI findings are always reviewed by experienced QA leaders. Explainability layers show exactly why issues were flagged, allowing humans to validate conclusions, provide edits, and make final quality decisions.
Test design – From weeks to hours
Generating test scenarios manually is one of the most time-consuming phases of the testing lifecycle. Our AI solutions ingest structured context – requirements, design documents, business process flows, user guides, wireframes and historical defects – to produce comprehensive and bias-reduced test cases and automation assets.
This has reduced greenfield test preparation effort by as much as 70%. A consumer electronics client compressed weeks of work into hours while significantly enhancing coverage.
Again, AI accelerates output, but humans provide authority. Test leads review all generated packs before adoption, perform sampling and risk-based validation, and adjust scope manually where products or regulatory environments demand deeper focus.
Automation stability – Self-healing scripts
Anyone running UI automation knows how fragile scripts can be. Minor front-end changes frequently break critical regression suites, eroding confidence in automation ROI.
AI now detects UI changes automatically and either heals scripts directly where confidence thresholds are met or proposes fixes along with confidence scores for human approval. This approach reduces false failures and dramatically lowers maintenance overhead.
Industry tooling shows up to 95% reductions in test-maintenance costs for UI-heavy test suites. One of our energy sector clients experienced a 50% reduction in automation maintenance effort.
Crucially, guardrails remain in place. Automated changes operate only within defined confidence limits. Every fix is logged and versioned and engineers retain control to approve, reject, or override any modification.
Result analysis – Faster, smarter insights
We apply multimodal AI to analyse test execution results alongside artefacts such as logs, videos, screenshots, and audio recordings. False positives filtered and root causes identified in seconds rather than hours.
QA teams spend less time triaging and more time focusing on real defects. Development receives clearer, faster defect routing, reducing wasted cycles and speeding resolution.
Human review again remains the final gate. AI proposes classifications and causes, but QA management validates all outcomes before routing defects or closing results. Explainability ensures accountability in every decision.
Why guardrails matter
Many organisations experiment with ‘DIY AI’, training testers as prompt engineers and encouraging tool-level experimentation. While this creates momentum, it rarely produces consistency or scale.
Without frameworks, businesses encounter unreliable outputs, governance gaps, and rising training costs, with no guarantee of reproducibility. Purpose-built frameworks create what ad-hoc tooling cannot:
Consistency through structured prompts and fine-tuned models
Validation via mandatory human checkpoints
Context through integration with tested systems of record
AI is an assistant, not a replacement
Every industrial shift has triggered job-loss fears, from mechanisation to automation and now AI. But in software engineering and testing, we believe that AI does something far more useful: it removes low-value repetition so experts can focus where human insight remains irreplaceable.
- Understanding product nuances
- Exploring edge cases that defy automation
- Making complex risk decisions that machines cannot
The winners won’t be organisations that apply the most AI. They’ll be the ones who know exactly where to apply it and more importantly, where it shouldn’t go.
The bottom line
AI in QE isn’t a silver bullet. It’s a powerful tool.
Used with discipline, it accelerates testing, increases coverage, and prevents defects before they escape into production.
Used without structure, it creates inefficiency, confusion, and dangerous overconfidence.
The outcome is entirely in your hands.
Coming next
In the next post of this series, Conor Thomson, Senior Solution Architect and AI innovation programme lead at Resillion, will dive deeper into our AI-driven test design framework and how we use generative AI to build context-aware, domain-specific test assets that save time and improve confidence.
Want to explore this in practice?
If you’re ready to see how structured AI can strengthen your QA process, get in touch. We’ll walk you through our frameworks, guardrails, and real-world results and help you move from experimentation to execution.
Robby Putzeys
Head of Quality Engineering
Our AI Partnering Strategy will help you navigate the AI landscape
Get in touch
Providing solutions in a unified, structured approach across all critical domains to enable end-to-end quality – Because we’re the only ones who can.
