A Chat With Adonis Celestine, Senior Director And Automation Practice Lead At Applause And AI45 Judge

Please introduce yourself and tell us about your role at Applause

I’m Adonis Celestine, Senior Director and Automation Practice Lead, at Applause, where I focus on helping organizations build scalable, intelligent quality engineering strategies for modern digital products and AI-driven systems. My role sits at the intersection of automation, AI, customer experience and trust, helping teams identify risks so products not only work technically, but also have a lower likelihood of unreliable or irresponsible behaviour in real-world conditions.

What does a Head of Automation actually do in the context of AI today?

The role has evolved significantly. Traditional automation was largely about validating predefined workflows and expected outcomes. In the AI era, we’re testing systems that are probabilistic, adaptive and sometimes unpredictable. That means we’re no longer just validating functionality, we’re evaluating behaviour, reasoning, safety, resilience and trustworthiness.

A large part of my work involves adversarial testing, prompt injection analysis, agentic AI validation, localization testing and assessing how AI systems behave across different users, cultures and environments. We also help organizations understand where automation ends and where human intelligence is still essential.

For those less familiar, how would you describe what Applause does and why testing and training AI models has become such a critical part of the development process?

Applause helps companies deliver better digital experiences through testing, quality engineering and actionable feedback from real people around the world. Our fully managed approach enables organizations to test under real-world conditions, across devices, locations and use cases – at the speed and scale required in the age of AI.

As AI becomes embedded into products and services, testing must go further than “does the feature work?” AI systems can hallucinate, misinterpret intent, generate harmful outputs or fail under edge cases that developers never anticipated. Training and testing AI models with diverse real-world inputs is now critical because these systems directly influence decisions, recommendations and customer experiences at scale. A model that performs well in a controlled lab environment may behave very differently when exposed to millions of unpredictable users.

Bias in AI is a major concern across industries, from finance to healthcare. Why is eliminating bias so difficult, and where do you see companies going wrong?

Bias is difficult to eliminate because AI systems learn from human-generated data and human history itself contains bias, imbalance and inequality. Even with good intentions, models can inherit patterns that reflect societal issues or incomplete datasets.

Where many companies go wrong is assuming bias is purely a technical problem. In reality, it is also a testing, governance and diversity problem. If your training data lacks representation, if your testing teams lack diversity, or if you only validate AI systems in ideal scenarios, bias will inevitably slip through. Another common mistake is treating fairness as a one-time compliance exercise rather than something that requires continuous monitoring and iteration.

Applause works with sectors like automotive, retail and banking. Are there any industries where AI bias or poor testing could have particularly serious consequences?

Healthcare and finance are particularly high-risk because AI decisions can directly impact people’s wellbeing, livelihoods and opportunities. A biased healthcare model could misdiagnose underrepresented populations, while biased financial systems could unfairly influence lending or insurance decisions.

Automotive is another critical industry, especially as autonomous and AI-assisted driving technologies advance. In these sectors, poor testing is not just a quality issue — it can become a safety, ethical and legal issue.

There’s a lot of focus on building bigger and more powerful models, but less on testing them. Do you think companies are underestimating the importance of validation and real-world testing?

Absolutely. There is currently a race to release increasingly capable models, but capability without validation can become extremely dangerous. Real-world environments are messy, adversarial and unpredictable. Models interact with ambiguous language, emotional users, malicious prompts and cultural nuances that are difficult to simulate internally.

Many organizations still approach AI testing with a traditional software mindset, but AI systems require continuous evaluation, red teaming and human oversight. The industry is gradually realizing that trust will become a competitive advantage — and trust only comes through rigorous validation.
 

What trends in AI are you personally most excited about right now, and which ones concern you the most?

I’m particularly excited about agentic AI systems and multimodal AI because they have the potential to dramatically improve productivity, accessibility and decision-making. We’re moving from AI that simply answers questions to AI that can reason, coordinate tasks and interact with the digital world in meaningful ways.

What concerns me most is the growing level of cognitive offloading and overreliance on AI systems without sufficient verification. There’s also increasing risk around adversarial manipulation, misinformation and autonomous systems making decisions that users may not fully understand. The technology is advancing faster than many organizations’ ability to govern or secure it responsibly.

As an expert judge on the TechRound AI45 judging panel for 2026, what are you most looking forward to in judging the competition?

I’m looking forward to seeing companies that solve meaningful real-world problems rather than simply adding AI as a marketing label. The most exciting innovations are often the ones that combine strong technical capability with genuine usability, responsibility and measurable impact.

It’s also fascinating to see how startups are approaching trust, explainability and human-AI collaboration, because those areas will become increasingly important over the next few years.
 

When reviewing entries, what will you be looking for? Is it technical innovation, real-world impact, responsible AI or something else entirely?

It’s a combination of all of those factors. Technical innovation is important, but innovation without practical value rarely succeeds long term. I’ll be looking closely at whether the solution addresses a real problem, whether it can scale responsibly and whether the team has genuinely considered trust, safety and user impact.

Responsible AI is no longer optional. The companies that stand out will be the ones that understand both the power and the responsibility that comes with deploying AI at scale.
 

What advice would you give to startups or founders entering AI45 who want to stand out in such a fast-moving and competitive space?

Focus less on hype and more on clarity of value. The AI space is crowded with companies making broad claims, but the strongest startups are usually the ones that deeply understand a specific customer problem and solve it exceptionally well.

I would also encourage founders to invest early in quality, testing and governance. Many startups treat those as problems to solve later, but trust becomes incredibly difficult to rebuild once it’s lost. The companies that will succeed long term are the ones building reliable, transparent and human-centered AI from the beginning.