Stratola Spectrum - Tech Conversations about AI, Data, and Automation

Everyone has been chasing smarter models. Bigger context windows. Faster inference. But the agents enterprises are deploying are still getting things wrong. Not because the models are not good enough. Because they do not understand the business.

In this episode of Stratola Spectrum, Dinesh Chandrasekhar sits down with Prukalpa Sankar, founder and co-CEO of Atlan, to talk about the problem sitting underneath almost every failed enterprise AI deployment right now:

• Why AI today is valuable but not useful and what the difference actually means.
• What the question "what are my top 10 customers" reveals about how unprepared most enterprises actually are.
• Why the bottleneck in enterprise AI is no longer technology and what is actually slowing deployments down.
• How Atlan went from five months in testing hell to five days in production for their customers.

🎙️ Prukalpa Sankar, Founder and Co-CEO, Atlan🎙️ Dinesh Chandrasekhar, Chief Analyst and Founder, Stratola🔗 https://www.stratola.com

00:00 — Introduction: Dinesh opens with the real question behind enterprise AI today.

00:23 — The Real Fault Line in Enterprise AI: Why the conversation is shifting from model performance to context.

02:13 — Meet Prukalpa Sankar, Atlan: Founder and Co-CEO of Atlan, co-founder of Social Corps the world's largest government data lake, Forbes 30 Under 30 and Fortune 40 Under 40.

03:40 — What Context Actually Is at a Systems Level: Why context is not just better retrieval or a larger token window.

04:09 — Why Onboarding an Agent Should Look Like Onboarding a Human: The parallel between how companies build institutional knowledge in human employees.

06:59 — AGI Is Already Here. It Is Just Not Useful: Why Prukalpa believes the intelligence problem is largely solved.

07:59 — The Five Layers of Context Explained: Using the question "what are my top 10 customers" to show why context is a stack of user intent, knowledge, meaning, semantics and data, not a single layer.

11:19 — Why Metadata Alone Was Never Going to Be Enough: How Atlan started as a data team solving its own problems.

13:41 — Why AI Agents Need Fundamentally Different Infrastructure: Every agent action starts with a context search and the latency.

15:11 — Atlan's Autonomous Workflow Milestone: The first week where autonomous workflows created more context than humans and AI-assisted workflows combined.

19:04 — Why Ontology Is Re-Emerging Right Now: The shift from bounded use cases like dashboards to unbounded agent use cases brought ontology back from niche construct to production requirement.

21:38 — We Are at the Worst AI Intelligence We Will Ever See: Why the current moment is the floor not the ceiling.

25:49 — The Difference Between Context for Data and Context for AI: Why having metadata, lineage and a business glossary is a necessary foundation but still not enough to take AI reliably into production.

32:31 — Why Shared Context Layers Collapse and How to Prevent It: The lesson from Looker.

35:19 — Centralized Platform, Federated Context: How Prukalpa thinks about ownership and governance architecture for context at scale across multiple agents and systems.

35:44 — MCP and Why Protocol Lock-In Is the Biggest Risk Right Now: Why the worst thing any leader can do today is commit to any single framework.

38:50 — Where a CIO Should Actually Start: Not with the context layer but with the top business problems that need solving in the next three to six months.

41:07 — From Five Months in Testing to Five Days in Production: Why the bottleneck in enterprise AI is no longer technology and what organizational change, ownership questions.

44:06 — What Breaks First When You Have 500 Agents: The three failure modes Prukalpa sees most often across enterprise AI deployments and how to build foundations that prevent each one.

47:11 — What Prukalpa Is Most Excited About in the Next 12 Months: The emotional journey organizations go through from fear to joy as they start giving AI more control.

95% accuracy on a medical exam. 20% accuracy in the real world. That gap is the healthcare AI story nobody is telling loudly enough.

In this episode of Stratola Spectrum, Dinesh Chandrasekhar sits down with Dr. Ron Razmi, ex-cardiologist, health AI investor at ZOI Capital and author of AI Doctor: The Rise of AI in Healthcare:

• Why the largest radiologist shortage in history exists despite years of AI investment.
• The pacemaker story. One missing data point. Almost a catastrophic decision.
• Why incumbents like Epic can kill your startup's pipeline without shipping a single feature.
• The B2C2B model quietly solving the enterprise sales problem in healthcare.
• Why 98% of digital health companies never reach an exit.

🎙️ Dr. Ron Razmi: https://www.linkedin.com/in/ronald-m-razmi-md-2b55b8

🎙️ Dinesh Chandrasekhar: https://www.linkedin.com/in/dineshc/

🔗 https://www.stratola.com

#HealthcareAI #AIinHealthcare #DigitalHealth #AmbientAI #HealthTech #MedicalAI #AI #StratolaSpectrum

00:00 — Introduction: Dinesh introduces Dr. Ron Razmi, his journey from practicing cardiologist to McKinsey consultant to health tech founder to author and investor in healthcare AI.

00:22 — Meet Dr. Ron Razmi: Ex-cardiologist, advisor at ZOI Capital and author of AI Doctor: The Rise of AI in Healthcare, written for builders, buyers and investors.

01:22 — The Biggest Misconception About AI in Healthcare: Everyone keeps saying AI will replace doctors.

03:21 — The Radiologist Shortage Nobody Is Talking About: AI has been detecting lesions for years.

07:27 — From WebMD to ChatGPT Health: WebMD was annoying. AI-powered self-diagnosis is a more serious problem.

09:34 — When LLM Accuracy Drops From 95% to 20%: Two studies from Nature Medicine and Microsoft confirm the same finding. Benchmark accuracy looks great. Real-world accuracy is a different story.

13:43 — Where AI Should Actually Start in Healthcare: Doctors do not need help with diagnosis. 95% of diagnoses are already on the chart. The case for starting with administrative burden.

16:19 — The Macro Environment Reshaping Healthcare AI Budgets: Medicaid rollbacks and ACA cuts are tightening budgets across the value chain and changing where spending is going.

19:03 — Why Selling Into Healthcare Takes So Long: Not one decision maker. The CMO, CTO, CFO and COO all need to say yes and each one has effective veto power.

21:52 — How Incumbents Block Startups Without Building Anything: Epic does not need a product. They just need to say they are working on it.

27:16 — The B2C2B Model Changing Healthcare Sales: Give it free to the doctors. Let them get dependent on it. Let them pressure the organization to buy it.

29:22 — Why Every AI Project in Healthcare Is a Data Project: The biggest barrier is not the algorithm. It is fragmented, siloed data that makes even good models unreliable in production.

31:07 — The Pacemaker Story: One Missing Data Point, One Near-Catastrophic Decision: Dr. Razmi's father. A beta blocker prescribed in Las Vegas.

34:38 — The Champagne That Went Flat: A startup celebrated getting oncology data from a major hospital. Then discovered the oncology team was on a completely separate EHR.

37:10 — The Adoption Scorecard for Builders, Buyers and Investors: The structured framework Dr. Razmi's team at ZOI Capital uses to evaluate health AI companies.

41:08 — Three Questions Every Hospital Should Ask Before Buying Clinical AI: Reimbursement, clinical evidence and liability. The questions most decision makers are skipping.

45:54 — The 2% Exit Rate Nobody Wants to Say Out Loud: 98% of digital health companies never reach an exit. Digital health is more than three times harder than the overall VC ecosystem.

48:52 — Where the Real Moat Is for Health AI Startups: Not the model. Workflow integration so seamless that people do not even realize their workflow changed.

51:57 — The Unit Economics Problem Most AI Startups Are Not Thinking About: Output tokens cost 20x more than input tokens.

The IoT promise was simple. Connect your machines. Get the data. Make better decisions. A decade later, most enterprises have the first two parts. The third one is still largely a human problem sitting on top of a very expensive data pipeline.

In this episode of Stratola Spectrum, Dinesh sits down with Dr. Jürgen Krämer, CPO and MD, Cumulocity, to talk about what it actually takes to close that loop in the AIoT era:

• How agentic AI is replacing dashboards and rule engines with digital workers that manage themselves.

• Why the gap between monitoring and operating is where most industrial AI value sits unclaimed.

• What a maintenance technician's job looks like before and after AI gets involved, and why the difference is striking.

• How digital twins evolved from 3D visualization into the contextual layer that makes AI decisions reliable.

• What state and context mean in a cyber-physical system and why most enterprises are still missing both.

No hype, just clarity.

🎙️ Guest: Jürgen Krämer, CPO and MD, Cumulocity, https://www.linkedin.com/in/juergenkraemer⁠🎙️ Host: Dinesh Chandrasekhar, Chief Analyst, Stratola, https://www.linkedin.com/in/dineshc/⁠ 📌 Chapters timestamped below - 00:00 — Introduction: Dinesh introduces Dr. Jürgen Krämer, his journey from founding RTM in 2007 through Software AG to leading Cumulocity today.00:15 — Meet Dr. Jürgen Krämer, Cumulocity: PhD from Marburg University, patent holder in complex event processing, veteran of industrial IoT since before it had a name.01:36 — What Cumulocity Actually Does: Real customer examples from ABB, Energon wind turbines and Eaton, and what connecting industrial assets at scale looks like in practice.03:10 — From M2M to IoT to AIoT: How Cumulocity evolved through three platform waves and what each transition demanded from customers.04:54 — Agentic AI Is Shifting SaaS From Tools You Use to Digital Workers You Manage: Why dashboards and rule engines are on their way out faster than most people expect.07:40 — Monitoring vs. Operating: The Line Most IoT Platforms Cannot Cross: Passive observation versus active management, with real wind turbine examples from Jürgen.10:27 — When Do You Automate and When Do You Keep a Human in the Loop: How a human-governed rule gets defined, approved and then automated at scale.14:15 — What State Actually Means in a Cyber-Physical System: Why sensor data alone is not enough and how metadata, asset hierarchies and semantic layers make AI decisions reliable.19:39 — Digital Twins Are Not 3D Models: The SAP partnership linking ERP master data to physical OT reality and why this bridge makes AI on industrial assets possible.25:38 — The Pump Four Moment: What AI-Assisted Field Service Actually Looks Like: Vague ticket in the old world. Diagnosis, part on order, repair guide and shutdown window in the new one.28:14 — You Cannot Trust the AI: Why human oversight in physical systems is not optional and what trustworthy AI actually requires in practice.31:01 — Edge vs. Cloud Intelligence: Latency, data sovereignty, air-gapped deployments and the develop-once-deploy-everywhere paradigm.34:38 — Intermittent Connectivity and Why Apache Pulsar Is Part of the Answer: The data broker approach that keeps industrial IoT reliable when connectivity is not.38:02 — Policy-Based Automation: How You Let AI Act Without Losing Control: Parameter changes, firmware updates, versioned and auditable human-approved policies.43:13 — The Competitive Landscape and Why Bob the Builder Is the Real Competitor: What happens when enterprises build their own IoT stacks and someone says "now do the global rollout."47:46 — The Moat for the Next Few Years: Semantic Layer on Top of Hyperscalers: Why buy and build gets stronger with agentic AI and what hyperscalers alone cannot provide.#AgenticAI #IoT #AIoT #EdgeAI #IndustrialAI #EnterpriseAI #DigitalTwin #AI #MLOps #StratolaSpectrum

AI adoption is not just a technology problem. It is a human one. And most enterprises are avoiding that conversation entirely.In this episode of Stratola Spectrum, Dinesh Chandrasekhar sits down with Kim Manis, Corporate Vice President of Microsoft Fabric Product at Microsoft, to unpack:✅Why AI amplifies organizational ambiguity instead of fixing it.✅How Microsoft Fabric and Fabric IQ are solving the semantic clarity problem at enterprise scale.✅Why governance treated as an afterthought in 2026 is a serious risk.✅What responsible AI adoption actually requires from the people inside the organization, not just the tools.✅The ROI conversation that enterprise leaders are still not having clearly enough.This is one of the most honest conversations on the human side of AI adoption you will find. No hype, just clarity.🎙️ Guest: Kim Manis, Corporate Vice President, Microsoft Fabric Product, Microsoft 🎙️ Host: Dinesh Chandrasekhar, Chief Analyst and Founder, Stratola📌 Chapters are timestamped below. Subscribe for more conversations at the intersection of Data, AI and the enterprise.🔗 Learn more about Stratola: https://www.stratola.com 🔗 Connect with Kim Manis: https://www.linkedin.com/in/kimmanis/ 🔗 Connect with Dinesh Chandrasekhar: https://www.linkedin.com/in/dineshc/

Time Stamps -

00:00 — Introduction: Dinesh opens with a framing that separates this episode from every other AI conversation, because the real story of every data platform shift was never about the technology.

00:42 — Meet Kim Manis, Microsoft Fabric: Corporate Vice President of Microsoft Fabric Product at Microsoft, 14 years in the data space, and the person running the platform that powers more than 30 million Power BI users worldwide.

01:19 — Why This AI Moment Is Structurally Different: AI does not just democratize access to data, it forces organizations to confront decades of implicit and unresolved meaning that they have been quietly avoiding.

04:19 — What Self-Service Analytics Actually Taught Us: Kim draws a direct line from the resistance to self-service BI a decade ago to what is happening with AI today, and the parallels are sharper than most people have acknowledged.

08:16 — AI Amplifies Organizational Ambiguity. It Does Not Fix It: Ask any organization for their top customers by revenue and watch the room collapse into a debate about definitions, because the problem was never the tool and it was always about meaning.

10:07 — How Microsoft Fabric Is Solving the Semantic Problem: Kim walks through the three-layer approach behind Fabric IQ, from centralizing data in OneLake to building organizational ontologies that give AI the context it needs to make the right decisions.

17:34 — Advice for Organizations Starting From Fear: Kim gives direct and practical guidance for smaller organizations that know AI is coming but are genuinely uncertain about where to start without creating more risk than they solve.

23:59 — Can Context Ever Be Fully Formalized: Dinesh asks the philosophical question at the heart of the episode and Kim's answer is honest and direct, because context is never done and that is exactly why the human layer in AI systems is not optional.

29:56 — 2026 Is the Year of Impatience: CDOs and CIOs are going to push prototypes into production this year whether the infrastructure is ready or not, and governance treated as hindsight is a real and coming problem for most enterprises.

35:02 — The Hardest Conversation Enterprise Leaders Are Still Avoiding: Kim describes the boardroom pressure to just put AI in the system, and Dinesh adds the piece nobody wants to say out loud about the ROI conversation that is largely missing from enterprise AI strategies right now.

Databases are no longer just storing data. They are being asked to reason over it. And most enterprises are not ready for what that means.In this season opener of Stratola Spectrum, Dinesh Chandrasekhar sits down with Shireesh Thota, Corporate Vice President, Databases at Microsoft, to unpack -

✅How AI is fundamentally rewiring database architecture.

✅The real story behind how OpenAI scaled Postgres to 800 million ChatGPT users.

✅Why RBAC is not built for autonomous agents.

✅What Microsoft is building next with Azure Horizon DB and Fabric.

This is one of the most technically grounded conversations on the future of enterprise data infrastructure you will find. No hype, just clarity.

🎙️ Guest: Shireesh Thota, Corporate Vice President, Databases, Microsoft

🎙️ Host: Dinesh Chandrasekhar, Chief Analyst and Founder, Stratola

📌 Chapters are timestamped below. Subscribe for more conversations at the intersection of Data, AI and the enterprise.

🔗 Learn more about Stratola: https://www.stratola.com

🔗 Connect with Shireesh Thota: https://www.linkedin.com/in/shireeshthota/

🔗 Connect with Dinesh Chandrasekhar: https://www.linkedin.com/in/dineshc/

#Databases #AgenticAI #VectorDatabase #MicrosoftAzure #DataArchitecture #CloudDatabase #EnterpriseAI #AI #MSFabric #CosmosDB #AzureDB

Timestamps -

00:11 — Introduction : Dinesh sets up why databases are one of the most important and underrated battlegrounds in the AI era today.

00:41 — Meet Shireesh Thota, Microsoft : Corporate Vice President leading all operational databases at Microsoft including SQL Server, Azure SQL, Cosmos DB, MySQL and Postgres.

02:04 — From Exact Lookups to Semantic Search : How AI has fundamentally changed the way data is queried and retrieved, and why vector search alone is still not the full answer.

06:08 — From Systems of Record to Systems of Reasoning : The biggest philosophical shift in database architecture today and what it demands from how data is stored and structured.

10:53 — Is SQL Actually Dead : The debate that refuses to go away gets a proper answer, and where natural language fits alongside SQL in the world of copilots and agents.

16:30 — How OpenAI Scaled Postgres to 800 Million Users : The real architecture behind ChatGPT, the engineering innovations Microsoft built to make it work, and the birth of Azure Horizon DB.

23:39 — Unifying the Entire Data Stack with Microsoft Fabric : Why the future is not more databases but one unified data estate, and how Fabric is Microsoft's long term bet to make that happen.

29:21 — Databases as the Memory Layer for AI Agents : The four types of agent memory explained and why every operational database is already a vector database.

36:10 — Rethinking Security and Governance for Agentic AI : Why traditional RBAC was never built for autonomous agents and what the next generation of access control actually needs to look like.

45:45 — Five Years from Now : What Should Microsoft Get Right: Shireesh closes with a grounded and honest vision for where databases need to be in a world where developers barely need to think about them anymore.

Bob Friday, Chief AI Officer,  @HPE  joins us as a special guest in this episode to discuss the role of a Chief AI Officer in today's world. Dinesh Chandrasekhar, Chief Analyst and Founder, Stratola, discusses relevant topics such as AI adoption, security and trust concerns, replacing humans with AI, Agentic AI's progress, and the future. With his vast industry experience, Bob takes us through his viewpoints from his vantage point as a #CAIO.#AgenticAI #AISecurity #NetworkAI #AIOps #StratolaSpectrum

Muckai Girish of Rockfish Data explains what synthetic data is and talks about its importance in powering the AI models of today. Dinesh Chandrasekhar, Chief Analyst, Stratola, digs up interesting insights from Girish on key concepts like trust, authenticity, security, data privacy, etc., related to #syntheticdata in this episode of #stratolaspectrum.

The problem of integration has evolved to be only more complex over the decades. As Agentic AI starts to address this problem, we find ourselves in need of agents having to communicate with each other. Check out this episode as Dinesh Chandrasekhar, Chief Analyst,  Stratola talks to the Chief AI Officer at Solace , Ed Funnekotter about the concept of #AgentMesh.#SolaceAgentMesh #agenticai #LLM #AI #GenAI #integration

With agentic platforms on the rise, one of the critical use cases that will be benefitted is enterprise workflow automation. This will be way beyond simple productivity workflows. This will include the integration of various enterprise applications. Will agentic automation complement or replace application integration? This is the area we explore in this episode.Our guest for this episode is Massimo Pezzini, a much-renowned ex-Gartner analyst who heads up research at  @Workato . Dinesh Chandrasekhar, Chief Analyst and CEO of  @stratolallc  talks to Massimo about various topics like -- Evolution of the automation space from BPM to RPA to now agentic automation- Composable applications- Adoption of agents across the industry- Agentic bubbles- Relevance of APIs in the agentic world- Agentic Integration Framework- Scope and relevance of SIs in the agentic world- Future opportunity space for agents#agenticai #rpa #automation #integration #appintegration #agenticautomation

Conor Twomey is a vocal AI advocate and evangelist. In his recent trip to Davos, he met with several seasoned tech dignitaries and luminaries. In this episode, he shares with Dinesh Chandrasekhar, Chief Analyst of  Stratola, those conversations, key themes, takeaways, and more. He delves into the topic of Agentic Software Engineering, where he foresees autonomous agents enabling every human being in this world to be an AI Engineer. The conversation then goes into topics of the viability and maturity of autonomous agents today and how they will evolve in the near future. Do NOT miss this episode, as there is so much to unpack that we left the tape rolling even past the half-hour mark.Guest profileConor Twomey - Accomplished executive leader with over 15 years of experience addressing the most demanding data challenges for top corporations worldwide. Conor is the former Head of AI Strategy at KX, a pioneer in real-time data analytics and decision intelligence. Under his leadership, KX successfully transitioned from a time-series database company to the Enterprise AI platform of choice for large-scale AI implementations. Before this role, Conor managed a 400-person organization encompassing Presales, Professional Services, Support, Managed Services, and Customer Success Management. Conor is a sought-after speaker and contributor, renowned for his insights on frontier technology topics including Data, Analytics, Machine Learning, AI, and Generative AI