David Colwell, Vice President of Artificial Intelligence and Machine Learning at Tricentis, brings a wealth of industry experience and academic curiosity to the DevReady Podcast. With a background that spans automation, quality engineering, and cutting-edge AI research, David leads the charge in making enterprise AI practical and effective. Having spent years building and training neural networks and raising five "natural networks" at home, David blends technical rigour with a wry sense of humour,...
David Colwell, Vice President of Artificial Intelligence and Machine Learning at Tricentis, brings a wealth of industry experience and academic curiosity to the DevReady Podcast. With a background that spans automation, quality engineering, and cutting-edge AI research, David leads the charge in making enterprise AI practical and effective. Having spent years building and training neural networks and raising five "natural networks" at home, David blends technical rigour with a wry sense of humour, offering a grounded and entertaining take on the challenges of AI adoption in business today. In this episode, David joins host Anthony Sapountzis to unpack the reality behind the AI hype and what it takes to build systems that truly deliver value.
The episode opens on a light-hearted note, with David comparing the training of AI models to parenting his children, drawing a parallel between machine learning hallucinations and childhood fibs. He shares his journey from automating his own QA role to spearheading Tricentis’ Vision AI product, reflecting on the early days of experimentation with models like BERT and GPT-2. David recalls how, in the past, it was a struggle to convince stakeholders that AI had practical utility, whereas now, the challenge lies in tempering the overwhelming demand for AI solutions with thoughtful implementation.
As the conversation deepens, David and Anthony discuss the growing trend of companies branding themselves as “AI startups” despite simply integrating off-the-shelf language models. David likens this to previous hype cycles around blockchain and cloud, pointing out that while we’re currently in a “quick wins” phase of AI, lasting value will come from building proprietary tools and tapping into unique data sets. He warns that early movers relying solely on public LLMs may struggle to differentiate, especially as customers become more agile in switching providers.
The duo also delve into the practical limitations of today’s AI tools, especially in areas where precision, repeatability, and auditability are non-negotiable. David argues that while hallucinations can be creatively useful (such as stress-testing systems with unpredictable input), they’re unacceptable in high-stakes environments like finance. He emphasises that any AI-powered system must be designed to tolerate and recover from failure, with humans in the loop where necessary. At Tricentis, generative AI is used to accelerate creative tasks like generating test ideas, but the execution remains deterministic to ensure reliability.
Closing the episode with wit and technical insight, David and Anthony critique the current obsession with AI agents. They explore how unnecessarily complex agent setups often lead to inefficiencies, especially when predictable outcomes could be achieved with simpler workflows. David amusingly describes his ongoing side project—testing how AI models choose between tools based on psychologically manipulative prompts, a concept he calls “SEO for tools.” It's a fitting end to a conversation that blends sharp analysis, ethical considerations, and a deep understanding of the balance between human and machine intelligence.
#ArtificialIntelligence #MachineLearning #AIStartups #TechPodcast #Automation #AIEthics #DevReadyPodcast
View more
