Disambiguation

In this episode of the Disambiguation podcast, host Michael Fauscette talks with Vasanth Mohan, Product Marketing Lead at SambaNova Systems, about why inference speed is the bottleneck holding back AI agents, how purpose-built silicon changes the economics of enterprise AI, and what business leaders need to understand about the infrastructure layer that sits beneath every AI application they use.Vasanth's career has tracked the frontier of emerging technology. He studied computer science at Stanford, worked in VR during the Oculus era, moved into edge computing and telco networking, and landed in AI infrastructure as the bottleneck shifted from software to silicon. At SambaNova, he leads technical product marketing and developer relations for the company's reconfigurable dataflow unit (RDU), a chip designed specifically for AI inference.The conversation covers which AI use cases are gaining real enterprise traction (coding is the base layer, coworker tools are actually coding under the hood, and the world is moving through three waves from chat to RAG to agents), what premium inference actually means for business leaders (not just faster tokens but the sweet spot of running large models fast enough for agents to work), why a coding agent that takes 25 hours could run in under one hour with the right infrastructure (and why that changes everything about productivity), how disaggregated inference splits the pipeline so that GPUs handle one phase while purpose-built chips handle another (prefill is compute-bound, decode is memory-bound, and they need different architectures), what the dataflow architecture does differently from GPUs (operator fusion that streams data through operations instead of bouncing back to memory), why speed is now priced as a distinct tier (and when paying double for fast tokens makes sense versus batching overnight), why JP Morgan Chase and sovereign AI deployments across Europe and Asia Pacific are choosing on-premises inference (regulatory requirements, audit concerns about closed providers, and the fact that 80% of data centers are air-cooled), what developers are building on SambaNova Cloud (planning agents on frontier models delegating to fast open source execution agents), and the one metric enterprise leaders should use to evaluate inference infrastructure (tokens per second per megawatt, normalized for the service levels their applications actually need).Timestamps:00:00 - Introduction00:32 - Episode title and guest intro00:49 - Vasanth's background: Stanford, VR, edge computing to AI infrastructure05:00 - AI agent use cases gaining real traction05:27 - Coding as the base use case for enterprise AI06:20 - Coworker tools are actually coding under the hood07:00 - Three waves: chat era, RAG deployments, agent era09:18 - Premium inference: a distinct product tier, not just speed10:18 - Model size versus speed: the competing forces11:52 - Prompt caching is solved; decode speed is the unsolved bottleneck12:14 - The 25-hour coding agent: what 20x faster inference changes13:57 - Two components of agent workloads: CPU execution and inference15:14 - Disaggregated inference: splitting prefill from decode18:30 - Dataflow architecture and operator fusion on the RDU20:09 - Speed priced as a tier: when paying double makes sense22:41 - Voice agents need sub-500 millisecond response time25:00 - On-prem inference: JP Morgan Chase and regulatory requirements26:17 - Open source models are essential for secure environments27:36 - Sovereign AI and geopolitical data concerns31:58 - Right-sizing models to the task34:00 - Developer patterns: planning agents plus fast execution agents35:44 - Evaluating infrastructure: tokens per second per megawatt37:06 - Six-month ROI for on-prem NeoCloud customers38:08 - Recommendation: Semi Analysis and Dylan PatelGuest: Vasanth Mohan, Product Marketing Lead, SambaNova SystemsHost: Michael Fauscette, CEO & Chief Analyst, Arion ResearchSubscribe so you never miss an episode.

In this episode of the Disambiguation podcast, host Michael Fauscette talks with Zoher Karu, Head of AI at Taelor, about why data governance has suddenly become a boardroom priority, how dirty data and missing business context cause AI projects to fail, and what practical steps leaders can take to build a data foundation that actually supports AI at scale.Zoher has spent his career at the intersection of data, analytics, and business strategy. He holds a PhD in electrical engineering from MIT, started at McKinsey, then founded startups in retail analytics and call center intelligence before leading enterprise-wide data and analytics organizations at Sears Holdings, eBay, Citibank (across 17 markets in Asia and Europe), and Blue Shield of California.The conversation covers why data governance is "sexy again" (AI amplifies data quality problems, so bad data now means bad decisions at machine speed), the blood-in-the-body analogy for enterprise data (every organ needs it, it should not be dirty or leaking), why a two-year data cleanup project is the wrong approach (clean as you go with a use-case-driven mindset), three reasons AI projects fail to deliver results (data quality and trust, missing business context that lives in people's heads not databases, and change management resistance), the "Susie knows how to do that" problem (business rules and institutional knowledge that AI agents cannot access), why pilot success does not predict production success (manually cleaned spreadsheets do not reflect real-world data), change management and the value exchange (people need to know what is in it for them), why productivity gains are not the same as business transformation (the real power of AI is reimagining processes entirely), cross-industry patterns in data challenges (siloed customer views, fractured definitions, departmental selfishness), how data teams are evolving toward full-stack roles with AI-assisted tools, governance as brakes that help you go faster (knowing the boundaries lets you push all the way to them), and practical first steps for business leaders (start with cost savings, ask employees what would make their job easier, build momentum through small wins).Timestamps:00:00 - Introduction00:32 - Episode title and guest intro00:45 - Zoher's background: MIT, McKinsey, startups, Sears, eBay, Citibank, Blue Shield03:18 - Business-first mindset for data leadership05:03 - Why data governance is sexy again06:18 - Data is like blood: the enterprise body analogy07:33 - When "revenue" means different things to different teams08:02 - Clean as you go, not a two-year cleanup project09:30 - The shift from AI productivity to AI governance10:00 - Three reasons AI projects fail10:55 - Missing business context: the "Susie knows" problem13:10 - Pilot versus production: the cleaned spreadsheet trap13:38 - Change management and the value exchange15:07 - Making existing work easier: the call center notes example15:28 - The real power of AI: reimagining processes entirely17:34 - Board-driven AI mandates and the fear of being left behind18:44 - AI is a tool, not a solution looking for a problem19:32 - Cross-industry patterns in siloed customer data21:49 - Citibank example: loan data missing time of day23:40 - Full-stack data teams and AI-assisted tools26:17 - Building governance that protects without becoming a bottleneck27:07 - Traffic laws analogy: rules of the road for AI29:32 - Brakes help you go faster30:14 - Practical first steps: start with cost savings and productivity32:06 - Ask your employees what would make their job easier33:04 - Success builds on success33:31 - Recommendation: Factfulness by Hans RoslingGuest: Zoher Karu, Head of AI, TaelorHost: Michael Fauscette, CEO & Chief Analyst, Arion ResearchSubscribe and turn on notifications so you never miss an episode.

In this episode of the Disambiguation podcast, host Michael Fauscette talks with Krisztian Kiraly, International Partnership Manager at OptiMonk, about why the future of e-commerce is AI-powered conversion rate optimization, how smart personalization turns existing traffic into significantly more revenue, and where the line sits between a relevant experience and a creepy one.Krisztian has spent more than a decade in digital marketing and e-commerce, with a deep focus on conversion rate optimization. He ran Conversion Masters, a CRO agency that worked with e-commerce brands to improve conversion rates and unit economics, before joining OptiMonk to lead international partnerships.The conversation covers the double squeeze facing e-commerce operators (ad costs up 222% in eight years while giants like Amazon, Temu, and Shein compete with nearly unlimited marketing budgets), why conversion before acquisition is the smarter growth strategy (pouring more water into a leaky bucket does not fix the leak), how AI-powered CRO removes bottlenecks that used to take developer teams weeks to address, the behavioral signals AI can read the moment a visitor arrives on a site, the evolution of exit intent from annoying pop-ups to dynamic AI-driven interventions tailored to the exact product page a visitor is leaving, why the product page is the new landing page (40 to 70% of visitors now land directly on product pages through Performance Max campaigns), AI-generated embedded content that shows benefit badges directly on product images, the 10 to 40% conversion rate uplift OptiMonk is seeing from A/B testing AI-optimized product pages, the privacy balance between first-party and zero-party data versus over-personalization that feels invasive, a 70% revenue increase for a large e-commerce store achieved through 100-plus small optimizations over one year, and practical first steps for e-commerce leaders who have not yet started using AI for conversion optimization.Timestamps:00:00 - Introduction00:32 - Episode title and guest intro00:48 - Krisztian's background: CRO agency to OptiMonk02:16 - The double squeeze: rising ad costs and giant competitors03:11 - Why increasing ad spend is not a long-term solution03:43 - AI-powered CRO: analyzing data and running tests at scale05:27 - How AI reads behavioral signals the moment a visitor arrives06:27 - The gap between AI-driven traffic and static landing pages07:25 - 67% of visitors now expect personalized experiences08:07 - Exit intent reimagined: dynamic headlines, images, and CTAs09:18 - Pop-ups are a tool, not inherently good or bad10:14 - Hyper-personalization powered by AI11:50 - Conversion before acquisition: the leaky bucket problem13:02 - CRO affects two profit factors versus one for ad spend14:34 - The product page is the new landing page15:57 - AI-generated embedded content and benefit badges17:23 - Answering "what is in it for me" in five seconds18:27 - 10 to 40% conversion rate uplift from A/B testing19:14 - Privacy and personalization: finding the golden balance21:39 - Zero-party data: the win-win of asking visitors directly23:03 - Where personalization crosses into creepy23:32 - 70% revenue increase: 100 small fixes, compounding results26:11 - Without data you are just another person with an opinion28:22 - First steps: know your data and start with product pages30:02 - Lifestyle images, benefit-focused headlines, and trust signals33:20 - Recommendation: Dan Martell on AI efficiency in businessGuest: Krisztian Kiraly, International Partnership Manager, OptiMonkHost: Michael Fauscette, CEO & Chief Analyst, Arion ResearchSubscribe and turn on notifications so you never miss an episode.

In this episode of the Disambiguation podcast, host Michael Fauscette talks with Hemang Upadhyay, Senior Product and AI Leader, about why most AI pilots die the moment they connect to real enterprise infrastructure, what it actually takes to deploy adaptive AI agents in production commerce environments, and how the convergence of AI and quantum computing will reshape enterprise architecture.Hemang has spent more than 16 years as a strategic product leader building AI-driven solutions for US commerce, e-commerce, and technology companies. His work across enterprise automation, AI-powered search, and recommendation engines has generated more than $450 million in combined business impact. He published a paper that won the IEEE Best Paper Award on quantum computing as a service, and has spoken at the AI Genetic Summit, Commerce Media Brand Summit, eTailer Boston, B2B Connect, and Identity Week America.The conversation covers the distinction between conversational AI and agentic AI (chatbots produce language, agents produce consequences), the five-layer pattern for scaling agents from pilot to production (constrain, translate, authorize, observe, recover), why autonomy should live inside the workflow but authority should live outside the model, the shift in commerce from helping customers find products to helping them achieve outcomes, the agent passport concept for identity and trust in agent-to-agent interactions, how mid-market retailers should adopt AI (data readiness first, then journey focus, then capability composition), six security controls for building trust into agentic architectures, the honest framing on AI and jobs (task automation, role redesign, and workforce decisions are related but not identical), and the hybrid future of quantum computing and AI.Timestamps:00:00 - Introduction00:32 - Episode title and guest intro00:48 - Hemang's background: 16+ years in product leadership and AI02:19 - Simplifying complexity across digital commerce04:45 - Research, speaking, and the IEEE best paper on quantum as a service07:05 - Conversational AI vs agentic AI: language vs consequences08:30 - Five components of an agent: goal, reasoning, memory, tools, authority10:00 - Connecting probabilistic decisions to deterministic systems12:42 - Scaling from pilot to production: the five-layer pattern13:45 - Constrain, translate, authorize, observe, recover17:36 - Autonomy inside the workflow, authority outside the model18:03 - Commerce agents: from product discovery to outcome achievement18:52 - Amazon Rufus to Alexa shopping transition20:06 - The apartment furnishing example: scattered work to one plan23:12 - Agent-to-agent identity: the agent passport concept25:18 - Authentication vs authorization at the moment of action28:00 - Mid-market retailers: start with focus, not scale29:25 - Data readiness: agents make bad decisions faster on bad data31:14 - Capability composition and graduated autonomy33:17 - Trust: does the system remain safe when the model is wrong?34:23 - Six practical controls for agentic security37:22 - Governance as executable policy inside the product39:00 - AI and jobs: three effects, not one narrative42:22 - Which parts of my work are becoming easier to automate?44:07 - Quantum computing and AI convergence: the hybrid future46:07 - Quantum as a service: orchestration across classical, AI, and quantum47:32 - Three actions for business leaders on quantum readiness50:03 - Recommendations: Ethan Mollick and Andrew NgGuest: Hemang Upadhyay, Senior Product & AI LeaderHost: Michael Fauscette, CEO & Chief Analyst, Arion ResearchSubscribe and turn on notifications so you never miss an episode.

In this episode of the Disambiguation podcast, host Michael Fauscette talks with Tim Lidman, Co-Founder and CEO of Clyde AI, about why the chat window is not the right interface for real human-AI collaboration, why most companies are stuck building Frankenstein workbenches of cobbled-together AI tools, and what it actually takes to design a product where humans and AI work together as a hybrid team.Tim's career traces the entire evolution of enterprise collaboration. He started in tech sales at WebEx and Cisco selling audio conferencing, moved to SAP SuccessFactors, then founded Think Tank, a structured collaboration platform based on decades of research into behavioral science and group decision support systems. Think Tank was acquired by Accenture in 2021, where Tim operated as a partner for four years. He co-founded Clyde AI with Chris Bricker to build what he calls AI-native collaboration: a product where AI is a first-class citizen in the architecture and UX, not bolted on top of legacy workflows.The conversation covers Tim's contrarian take on LLMs and the promise of natural language interfaces (prompt engineering and context engineering are just new paradigms users have to learn, not the elimination of paradigms), why single-threaded chat is the wrong model for complex problem solving, the AI advisor concept (multi-threaded AI with separate domain knowledge and personas working in parallel without polluting each other's context), the design blueprint of mapping what AI is good at against what humans are good at, why humans are still better at inventing new information and making judgment calls, the Frankenstein workbench problem (employees cobbling together tools just to check the AI box), fear-driven adoption as terrible leadership, the micro win approach (20 minutes to solve one small problem and build trust), why anyone using AI for more than 10% of their workflow is ahead of 99.99% of workers, why the term change management may not survive (change is now constant, not a project with a start and end date), the people-process-technology split that must blend into one unified experience, and the industrial revolution parallel compressed 100x.Timestamps:00:00 - Introduction00:31 - Episode title and guest intro00:50 - Tim's background: heavy metal drummer to tech sales to collaboration to AI02:07 - The golden thread: evolution of collaboration from audio conferencing to AI03:15 - Enterprise social networking: Microsoft buys Yammer for $1.2 billion03:42 - Think Tank: structured collaboration and group decision support systems05:32 - The hypothesis for Clyde: AI removing expensive synthesis work06:41 - What AI cannot replace: tribal knowledge, human judgment, buy-in08:13 - Building Clyde: AI where AI excels, human UX where humans excel09:33 - AI advisors: multi-threaded thinking inside a collaborative workspace11:14 - The chat window problem: LLMs created a new paradigm, not eliminated one14:03 - Putting the onus on the tool to extract context and intent16:19 - Single-threaded chat vs multi-threaded problem solving18:37 - Example: business plan with parallel risk, strategy, and financial advisors20:29 - Legacy tools bolting AI on top of old architectures22:32 - Fear-driven adoption: use AI or you are fired29:10 - Building hybrid teams: human-to-human vs human-to-AI collaboration30:52 - Humans can invent new information; AI is pattern recognition34:38 - Change is now constant, not a project with start and end dates36:28 - The term change management may not survive38:32 - Continuous optimization: building and optimizing happen simultaneously41:42 - People-process-technology must blend into one unified experience44:11 - The industrial revolution parallel, compressed 100x46:27 - Recommendation: Sapiens by Yuval Harari47:20 - Recommendation: CEO of Lovable, 0 to $100M ARR in nine monthsGuest: Tim Lidman, Co-Founder & CEO, Clyde AIHost: Michael Fauscette, CEO & Chief Analyst, Arion Research