Show Notes
In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Antonio Mallia, founder and CEO of Seltz.ai, to explore one of the most overlooked infrastructure questions in the AI era: what happens to web search when the user is no longer a human? Antonio opens with a career that runs from the University of Pisa's information retrieval tradition through a PhD at NYU to research roles at Amazon, Pinecone, and Bloomberg — and traces the moment the idea for Seltz crystallized while building Alexa's question answering engine, when it became clear that the next consumer of search results was not a person but a machine. He then explains why Seltz took the harder path of building its own full retrieval stack from scratch rather than wrapping Google or Bing like most AI search products — and why that distinction matters deeply. Traditional search engines were optimized for human patience: ranked lists, short snippets, spell correction, ambiguous queries refined over multiple attempts. LLMs need something fundamentally different — full paragraphs without truncation, long and specific queries that nail the topic on the first attempt, and above all, freshness, because the one thing a model cannot learn in training is what happened after its cutoff date. Seltz also focuses its index on the content LLMs struggle most with — tables, financial records, SEC filings, and structured data that traditional crawlers skip and models cannot reliably memorize. Antonio closes with a prediction that reframes the entire stakes of what he is building: human search queries have plateaued, but machine queries are growing exponentially, and as people spawn multiple specialized agents to handle their knowledge work, the query load on web infrastructure will reach a scale never seen before. The companies that build for that future — with efficiency and freshness at their core — will define the next twenty years of the web.
If you liked what you heard today, please leave us a review - Apple or Spotify.
Learn more about your ad choices. Visit megaphone.fm/adchoices
More from The Digital Executive
In this episode of The Digital Executive Podcast, host Brian Thomas speaks with Chandan Mahajan, co-founder and Chief Growth Officer at DotConnect, about helping enterprises turn AI ambition into measurable business results. Drawing on more than 20 years of enterprise technology experience, Chandan explains why industry-specific AI can outperform general-purpose solutions by leveraging the unique context, workflows, and expertise of a business. He shares how DotConnect combines consulting, engineering, and its Compass Studio platform to accelerate AI implementation while allowing companies to retain ownership of their solutions and data. Chandan also explores the future of agentic AI, emphasizing that organizations must become AI-native while remaining human-owned. His central message: companies can rent AI models, but they must own their technology, business context, and capabilities to maintain control and long-term competitive advantage.
In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Dharmateja Uddandarao, Senior Data Scientist and Statistician at Amazon, to make a distinction that most business leaders have never fully grasped — and that costs organizations millions of dollars in misattributed decisions every year. Dharmateja opens by tracing how causality became the through line of his career not by design, but by repeatedly hitting the same wall: a beautiful, high-accuracy model would ship a recommendation, a leader would ask whether the product actually caused the outcome or whether it would have happened anyway, and the model had no answer. That gap — between what happened and what would have happened — is where statistics, data science, and economics all converge, and it is the problem Dharmateja has spent his career learning to solve. He then breaks down why all three disciplines are necessary: classical econometrics has the deepest theory of identification but is often rigid; machine learning handles complexity and scale but is causally blind, happily telling you that ice cream sales cause drownings because both rise in summer; and causal inference theory provides the connective tissue — the language of counterfactuals and potential outcomes — that makes it possible to ask what the world would have looked like if you had acted differently. He illustrates this with an ad measurement example where a naive eleven-point lift becomes a credible eight-point lift once confounding is properly addressed, and walks through counterfactual forecasting as a method for estimating causal effects when no randomized experiment was possible. His close is direct: the next decade of AI is not about prediction becoming cheaper — it already is. It is about decisions becoming more intelligent. And that requires causal grounding, honest uncertainty, and economic context, not just bigger models.
In this episode of The Digital Executive Podcast, host Brian Thomas speaks with Shauna Geraghty, Senior Vice President and Global Head of People & Talent at Talkdesk, about how AI is transforming HR, workforce strategy, and the future of work. As one of Talkdesk’s first U.S. employees, Shauna helped scale the company from a startup to more than 1,500 employees. She explains why HR and IT must work together to redesign how organizations operate, using AI to automate repetitive tasks while empowering employees to focus on strategic work and meaningful relationships. Looking ahead, Shauna shares three essential capabilities for future HR leaders: business and technology fluency, organizational and workforce architecture, and transformation leadership. Her message is clear: successful AI adoption is about more than efficiency—it’s about amplifying human potential and creating greater business value.
In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Seymour Segnit, founder and CEO of MagFast, to explore what it takes to build a hardware company around a single, deceptively simple idea: that charging your devices should be effortless, elegant, and yes, delightful. Seymour opens with a career that spans Oxford engineering, Ogilvy and Saatchi advertising, Capital Radio, early SaaS, and USB charging — a patchwork of experiences that he argues has made him a better operator because he has enough fluency in each discipline to actually understand the people working across them. MagFast’s core insight is that people do not need more adapters. They need one integrated system — chargers, cables, and power banks that share a common design language and snap together magnetically using the industry-standard Qi2 and USB-C technologies that now span both Apple and Android ecosystems. By locking into those standards rather than proprietary ones, MagFast can offer genuine interoperability without asking customers to choose a walled garden. He then walks through what twenty-five million dollars in direct-to-consumer sales has taught him that a traditional hardware company never learns: you cannot know your customers if you do not know who they are. Building a unified customer data layer has allowed MagFast to deploy a voice agent named Maggie that actually delights people — not because it sounds good, but because it knows that your package was delivered to your porch in Kansas City at 2:30 yesterday afternoon. The difference between a great voice and a great agent, Seymour notes, is the data behind it. He closes with a vision of where charging is heading — wireless magnetic connections becoming as routine as Wi-Fi, the USB-C standard lasting decades — and a business philosophy built around the toughest standard he knows: is it delightful? If not, it is not done.
In this episode of The Digital Executive Podcast, host Brian Thomas sits down with Jake Sylvester, founder of Hypercall, to explore what happens when a builder who started with a high school cybersecurity company, got pulled into Bitcoin through ransomware response, and spent years handling tens of billions in cross-chain volume at Synapse Protocol turns his attention to one of the hardest unsolved problems in crypto: on-chain options liquidity. Jake opens by pushing back on the industry’s long-held assumption that retail investors cannot handle options, pointing to the millions of Robinhood users — chefs, Uber drivers, everyday people — who already trade options and understand them well. The gatekeeping, he argues, has more to do with tradition and misplaced paternalism than it does with actual investor protection, and academics like Nassim Taleb have long questioned whether the volatility models underpinning that gatekeeping are even correct. Hypercall’s answer is to build options for everyone — crypto and equities, twenty-four hours a day, seven days a week, including nights and weekends when underlying markets are closed. He explains the engineering challenge behind that: traditional options market-making firms have spent decades running delta-one desks across time zones just to approximate weekend S&P hedging, and Hyperliquid’s liquid on-chain venue has made that tractable at a scale that would have seemed impossible even a few years ago. A single S&P 500 options trade on Hypercall became fifteen percent of Hyperliquid’s S&P volume — a signal of just how much latent demand exists for this kind of access. He closes with a clear-eyed look at what still needs to be built: better tokenization of equities on-chain, cleaner handling of dividends and stock splits in a system that can never pause, and continued progress on the regulatory front. His five to ten year view is that the majority of options trading moves on-chain entirely — and that the people who benefit most will be the ones who were never allowed in before.