Why does Gartner report that 85% of AI projects fail?
According to Gartner, 85% of AI projects fail not for technology reasons, but for reasons of technology alignment to customer needs and business needs.
Podcast · Tech & Cybersécurité
Mark hosts conversations with industry leaders on how AI fundamentally reshapes product design, team structures, and UX strategy for the next generation of digital products.
AI is not a feature—it is a complete restructuring of how products get built, who builds them, and what interfaces look like. This podcast cuts through the hype to surface the real operational shifts happening in UX, product strategy, and startup design. Conversations with Dan Saffer, Luke Wroblewski, Barry O'Reilly, and others reveal that prompt boxes are already obsolete, design systems must become generative, and product teams need to unlearn legacy workflows entirely. The next 3–5 years will separate teams that treat AI as a tooling upgrade from those that redesign their entire operating model.
Explore all episodes of AI, Product and Design Podcast on Listenly
This is not a surface-level exploration of AI capabilities. The podcast addresses the structural problems that emerge when organizations try to retrofit AI into existing products and workflows. Mark and his guests dissect why the prompt box interface was broken from the start, how design systems must evolve to support generative outputs, and what happens when AI agents make decisions autonomously. Recurring themes include the collapse of traditional UX hierarchies, the need to rethink data architecture for AI workloads, and the strategic shifts that distinguish winners from those left behind. The conversations assume listeners already understand basic AI concepts and need to know how to actually build and ship products in this new reality.
Product managers trying to navigate AI adoption without repeating the mistakes of 1,000 startups need these frameworks. Designers grappling with how to maintain craft and intentionality when systems generate outputs autonomously will find concrete patterns here. Founders and engineering leaders confronting organizational restructuring around AI should listen before making irreversible decisions about team composition and workflow redesign.
The episode titles themselves encode a coherent thesis: AI won't fix bad UX, interfaces are dissolving, most AI projects fail for human reasons not technical ones, and the future belongs to teams building "boring" AI products that solve real workflows. This is not the podcast equivalent of "AI will change everything"—it is the podcast that explains specifically how, why, and what to do about it. Episodes dive into venture studio models that escape the unicorn trap, the role of AI in transforming raw data into compelling user experiences, and the precise evolution of design thinking required to remain relevant.
Organizations will shift from hiring specialized single-discipline roles to seeking hybrid product builders who can code, design, understand AI, and think systemically about workflows. Design tools, project management systems, and product development methodologies will be rebuilt around agentic patterns. The competitive advantage moves from "we have AI" to "we rebuilt our entire organization around how AI actually changes workflows." Teams that simply add ChatGPT to Figma and call it a day will lose to teams that reimagine what product means when intelligent systems collaborate with humans autonomously.
AI's impact on product is not incremental—it is a reset of the rules by which products are architected, teams are organized, and interfaces evolve. Organizations treating it as a feature upgrade will be outpaced by those rebuilding from fundamentals.
Listen to all episodes and gain the frameworks you need
Subscribe now to stay ahead of the AI product revolution
Most teams treat AI as a feature rather than a fundamental shift in how workflows operate. Without redesigning core processes, integrating AI into existing systems creates friction, not value. Success requires rethinking the entire product strategy, not bolting AI onto legacy interfaces.
AI-native interfaces are moving beyond text input to agent-based workflows where users define intent once and systems execute autonomously across multiple tools. Agentic design treats AI not as a tool but as a collaborative entity with persistent state and contextual understanding.
The next generation requires hybrid skills: designers must understand AI workflow architecture, PMs must grasp UX implications of agent behavior, and engineers must collaborate on prompt crafting and output validation. Siloed roles break down when AI capability reshapes what "product" means.
Traditional design systems based on static components are giving way to generative design systems where AI produces variations, manages state across workflows, and adapts interfaces in real time. Tool interoperability and AI-to-AI communication replace manual handoffs between design, dev, and product.
6 questions answered · 1 episode indexed
According to Gartner, 85% of AI projects fail not for technology reasons, but for reasons of technology alignment to customer needs and business needs.
Snowball Sprint is an extension of the old design sprint concept that starts with framing the problem before writing any code, using use case, data, and value exercises.
The value matrix, authored by Orjansson Gupta who created Salesforce Einstein, assigns cost or benefit to each outcome (true positive, false positive, true negative, false negative) and multiplies that by how often the outcome actually occurs.
The wrong use case is the primary reason that torpedoes the majority of AI projects. Teams must understand what problem they are solving, what the job to be done is, what pain they are addressing, how it is currently being done, and what AI will bring that is not being done today.
Greg has created the UX4AI Professional Certification program based on his book UX4AI, which teaches designers and product managers how to apply fundamental UX skills to AI projects.
Greg compares current AI adoption to the Hadoop syndrome, where boards demanded AI without understanding what to use it for, just like they demanded Hadoop when it first came out without knowing its purpose.