Future Proof: Building AI Products that Last

How do you build AI for industries where "average" is a failure? In this episode of Future Proof, Morten Bruun, Member of Technical Staff and Product Manager at Hebbia, breaks down why off-the-shelf models like Claude and ChatGPT aren't enough for the high-stakes worlds of finance and legal.Morten shares his journey from co-founding Flash Docs (acquired by Happier) to developing a platform that manages terabytes of unstructured data for leading investment banks and law firms. We dive deep into:The "Last Mile" Challenge: Why 90% accuracy is effectively 0% for professional standards.Process Engineering: Moving beyond chatbots to create a "canvas" for team-specific workflows.The SaaS-pocalypse: How software companies can differentiate as the cost of code drops to zero.The Technical PM: How AI is compressing feedback loops and turning product managers into builders.Whether you’re a founder navigating the "SaaS-pocalypse" or a product leader integrating AI, Morten’s insights on winning trust in regulated industries are essential.Ready to scale your product’s integrations without the engineering overhead? Visit UseParagon to learn how to ship production-ready integrations in days, not months.Chapters00:00 - Meet Morten Bruun and Hebbia00:42 - From Flash Docs to Acquisition: Morten’s Journey03:15 - Why Finance Needs an Edge (The Problem with "Average" AI)05:18 - Process Engineering: Moving Beyond the 30,000ft View08:02 - The Retrieval Gap: Analyzing Terabytes vs. 20 Documents11:19 - Solving the "Last Mile" for Professional Standards14:19 - High-Stakes Outcomes: Why Precision Matters17:16 - The "Canvas" Model and AI Strategists21:54 - Surviving the SaaS-pocalypse: Differentiation in the AI Age25:49 - The Changing Role of Product Management and Technical Staff36:01 - Winning Trust in Highly Regulated Industries40:43 - Customer Empathy: Focusing on the "Job to be Done"

In the latest episode of Future Proof, we sat down with Kevin Wang, Chief Product Officer at Braze, to unpack how AI is reshaping the relationship between brands and consumers and what marketers need to do to stay ahead of the shift.We cover:Why the next big ad surface isn't search or social and how LLMs will know what you want before you askHow personalization is evolving from coarse-grained categories to something closer to music taste than ice cream flavorWhy the real battle in AI-driven commerce isn't capability, it's who owns the consumer contextHow Braze is using reinforcement learning and in-product agents to give marketers exponential strategic leverageWhy human-in-the-loop isn't going away and how the role of the marketer is shifting up the stackWhat every major platform shift tells us about where brand-to-consumer relationships are headed nextHighlights: 00:00 — AI as the Next Big Ad Surface: Why LLMs Are Opening Up a Platform Shift Bigger Than Social04:30 — Spelling, Diction & Depth: How Agents Will Personalize Beyond What Facebook and Google Can See09:15 — Who Owns Your Preferences? The Battle Between Foundational Models, Retailers, and Consumers14:00 — Agentic Shopping and the Limits of Automation: Why People Still Want to Buy for Themselves19:30 — Braze's AI Decisioning Studio: Reinforcement Learning, Guardrails, and Offer Fit25:00 — Human-in-the-Loop: Why AI Is Also Really Good at Checking on Humans30:00 — Enterprise AI Adoption: AI Academies, On-Ramps, and Getting Comfortable with Experimentation36:00 — LLMs Are Not Deterministic: What Happens When Your Model Picks Its Own Magic Numbers41:00 — The Future of Brand-to-Consumer Relationships: More Personal, Not More DistantShip integrations 7x fater with Paragon.useparagon.com/?utm_source=spotify&utm_medium=organic_social

In the latest episode of Future Proof, we sat down with Elise Gonzalez, Staff PM at Databricks, to unpack how Agent Bricks and the MCP marketplace are shaping what enterprise-ready agents actually look like.We cover:Why governance, audit-ability, and user-level identity are the real blockers to enterprise agent adoptionHow MCP unlocked momentum — and why marketplaces, validation, and control are required to make it safeWhen multi-agent architectures make sense (and when they just add complexity)Why most MCP servers today won’t survive long-term — and what agent-native tools need to look likeHow AI is changing who uses data platforms, not just what they buildHighlights(00:00) Multi-Agent Architecture From Microservices to Agent Supervisors(02:12) Agent Bricks Overview Core Use Cases Extraction, Knowledge Assistants, Composition(04:01) MCP Marketplace Discovery, Governance and Enterprise Control(07:39) Validating MCP Servers Auth, Permissions and Auditability(11:37) Preventing Data Leakage Tool-Level Controls and Safety Patterns(14:58) MCP vs REST Rebuilding APIs for Agent-Native Workflows(18:48) Multi-Agent Risks Privilege Escalation, Context Sharing and Debugging(22:16) When to Use Multi-Agent vs Single-Agent Architectures(30:13) Why Databricks Invested in AI Agents and Data Governance(33:22) The Shift in Users Engineers, Business Users and Databricks OneShip integrations 10x faster with Paragon

In the latest episode of Future Proof, we sat down with Christine Zdelar, Director of Product at Intercom, to unpack how Fin has evolved from one of the first AI CX agents in production into a best-in-class support agent.We cover:- The Fin Flywheel and why training, testing, and analysis are core to agent success- Why Intercom went all-in on outcome-based pricing- How MCP, permissions, and oversight show up in real-world production systems- What it takes to push AI resolution rates from ~20% to 60%- And why Intercom is investing deeply down the stack, including building its own models Timestamps:(0:00) Clips(0:55) Introduction(1:26) Getting Fin production ready(4:46) Adopting MCP, is it ready?(7:26) Integrating with helpdesks(11:26) Connecting with knowledge bases(13:26) Resolution-based pricing strategy(19:26) Is the market ready for end-to-end AI resolutions?(23:26) Intercom adopting AI processes(25:26) What needs to happen for customers to adopt Fin(28:26) Intercom builds its own models?(34:26) Closing thoughtsShip integrations 7x faster useparagon.comHost: Ethan LeeGuest: Christine ZdelarProduced by: Forrest Herlick

In Episode 10 of Future Proof, we sit down with Noaa Ilani, VP of Product at Gong, to unpack how AI is transforming sales, Gong’s vision for agentic workflows, and what the future of human–AI collaboration really looks like.(0:00) Clips + Introduction(1:50) Fully automated sales rep?(5:00) Sales is built on human trust(5:30) Solving the memory problem(10:00) Understanding conversations(13:00) Accounting for tone(16:20) Embedding agents into other systems? Gong’s vision (22:00) Integrating with other AI tools(23:00) Mitigating missing data(28:35) Human collaboration with AI(34:00) Closing thoughts

In the latest episode of Future Proof, building AI products that last, Mike Gozzo, Chief Product and Technology Officer at Ada, shares how Ada is helping enterprises move from scripted chatbots to reasoning-driven AI agents, why trust and human empathy remain the foundation of customer experience, and how metrics like “automated resolution” are redefining success in enterprise AI deployments.Highlights:0:00 Clips0:36 Intro1:54 Future of customer experience 5:00 Building trust 8:00 Getting executive buy-in10:00 What true resolution looks like13:00 How Ada measures success with “automated resolution”16:00 Mirroring human QA processes18:00 Why domain expertise is the new superpower for AI founders19:30 Grounding responses with retrieval, context, and memory22:00 Balancing latency, reasoning, and customer experience24:00 Deploying agents safely and effectively26:00 Voice automation and ROI across industries29:00 Surpassing human agents and the future of AI-driven CXShip integrations 7x faster: https://www.useparagon.com/Host: Ethan LeeGuest: Mike GozzoProduced by: Forrest Herlick

In the latest episode of Future Proof, building AI products that last, we chat with Elliott Choi, Director of Product at Cohere. Elliott shares how Cohere is shaping the path from simple demos to enterprise deployment, why oversight, auditability, and security are essential for agent adoption, and the role of automation and memory in moving toward fully autonomous AI agents.Highlights:(0:00) Clips(1:07) Intro & $500M fundraise at Cohere(2:20) Defining AI agents for the enterprise(5:00) Traceability, observability & automation guardrails(8:12) Automating well-defined enterprise processes(11:21) Why flashy demos break in enterprise(12:09) Architecture pitfalls: prompts, data sources & security(15:15) Managing context & RAG performance(17:08) Building semantic vs. procedural memory(19:24) Permissioning & secure data access with SSO(21:32) Why memory makes agents sticky(25:54) From automations to fully autonomous agents(26:57) Closing thoughtsShip integrations 7x faster → https://www.useparagon.com/Host: Ethan LeeGuest: Elliott ChoiProduced by: Forrest Herlick

In the latest episode of Future Proof, building AI products that last, we chat with Megh Gautam, Chief Product Officer at Crunchbase. Megh shares how Crunchbase's customer-first approach to AI implementations preserved user trust while transforming search workflows, and practical strategies for rolling out AI features without alienating existing power users.Highlights:(0:00) Intro(1:00) Crunchbase positioning as role of CPO(3:40) Adding AI capabilities to current product(7:15) Learnings from experiments that didn’t work out the way it was expected(10:45) Metrics used to measure AI features delivering value(14:00) How Crunchbase thinks about aligning with user needs, not just cool tech(20:30) Where are AI builders making mistakes between AI technology and their users(24:00) Navigating the prioritization between core platform improvements and AI capabilities(28:30) Closing thoughtsShip integrations 7x faster www.useparagon.comProduced by: Forrest HerlickHost: Ethan LeeGuest: Megh Gautam

In today’s episode of Future Proof, we sit down with Anant Bhardwaj, Founder & CEO of Instabase. Anant shares how Instabase pioneered the use of transformer models for unstructured enterprise data, and how the rise of large language models is shifting value creation from field-level accuracy to full end-to-end workflows.We discuss:Finding product-market fit with nascent AI technologiesMoving beyond “LLM wrappers” to deliver end-to-end customer valueLessons from building in regulated industries like banking and healthcareWhy user experience, not model quality, defines AI product successHighlights:(0:00) Clips(0:53) Opening thoughts(2:15) Instabase early journey before the AI wave(5:28) How new models have shifted product development strategy(10:16) Creating value for customers with LLMs(12:06) Working with heavily regulated customers(15:16) What’s been the hardest about building infrastructure for unstructured data use cases?(21:06) AI infrastructure bets companies should be making today?(23:06) How Agents will work with unstructured data?Host: Ethan Lee https://www.linkedin.com/in/thatethanlee/Producer: Forrest Herlick https://www.linkedin.com/in/forrestherlick/Guest: Anant Bhardwaj https://www.linkedin.com/in/anantpb/Ship integrations 7x faster -> www.useparagon.comLearn more about Instabase -> https://instabase.com/

In today’s episode of Future Proof, we sit down with Laura Burkhauser, VP of Product at Descript. Laura shares how Descript’s design philosophy enables users to maintain control over AI creative workflows, and practical strategies for building AI interfaces that feel collaborative rather than restrictive.We discuss:Designing user interfaces in AI productsHow to make AI feel naturalHow to give users a sense of controlHighlights(0:00) Clips(0:37) Introduction(1:00) Designing interfaces that make users trust they’re in control(6:00) Open World AI Design Philosophy(8:40) Was there ever a tool you needed to modify once AI got access?(13:45) How to encourage users to understand what AI is good at(19:00) Giving control back to users(20:30) How to handle latency(23:00) Context switching while AI does its work(25:00) Biggest UI opportunities(30:00) Will we progress beyond a chat model?

In this episode of Future Proof, we sit down with Aman Khan, the Head of Product at Arize AI. Aman reveals why traditional product metrics fail for AI systems and shares Arize's framework for building evaluation systems that actually predict real-world AI performance, plus the emerging PM skills that separate successful AI products from failed experiments.We discuss: How AI builders should think about evaluationsThe role of the AI pm and how product management is evolvingHow you should build with the expectations of foundation models changing.(0:00) Highlights(0:37) Intro(1:40) What is an AI pm(4:10) How PMs are evolving with AI(8:10) The Aha moment in AI(11:50) What AI builders should think about evaluations(19:40) How AI builders best leverage their time in AI evaluations(23:40) Prompt iteration - if your evaluations are not ideal, how do you iterate?(27:40) What’s the minimum viable eval someone should write(30:40) How would prioritization change based on the future of AI models(36:40) Final thoughts(38:00) Ethan's reflectionShip integrations 7x faster https://www.useparagon.com/Watch all Future Proof episodes: https://www.useparagon.com/future-proof

In this episode of Future Proof - Building AI products that last, we sit down with WorkOS founder and CEO, Michael Grinich, on what it takes to make MCPs production-ready.We discuss:What it means to be ‘MC Pilled’What’s missing from MCP to make it enterprise-ready?How to build a product for a world of agentic usersHighlights(0:00) Clips(0:44) intro(1:11) MCP night(3:49) What is preventing demos from production ready AI agents(6:41 ) Authorizing tools in MCP(9:01) Should AI builders develop protocols themselves?(11:11) When should you build a MCP server(12:41) MCP ecommerce shop(14:05) Why are permissions a problem in rag(22:11) WorkOS acquisition of warrant (FGA)(26:11) How auth changes once agents become common users of webapps (oAuth 3.0)(30:11) How should builders shape their apps to be prepped for the future?Scale your AI product's integration roadmap with Paragon: https://www.useparagon.com/?utm_source=futureproofpodcastSign up for their MCP night → https://workos.com/mcp-night

Welcome to the second episode of the Future Proof Podcast: Building AI Products That Last! In this episode, we chat with Richard Socher, CEO and Co-founder of You.com, about how they pivoted from consumer search to enterprise, and now outperform OpenAI.We cover:Building for the future of Enterprise productivityHow to build a competitive advantage against industry leaders like Open AIIs chat the final interface for work?(00:00) Clips(00:50) Intro(01:10) How Richard started You.com(05:40) Building for the enterprise use case(09:20) How to outperform OpenAI(14:00) Advice for AI founders(19:00) Is chat the final interface for work?(23:30) Will there be a consolidation of AI?(24:40) How will AGI affect the future of work?Scale your AI product's integration roadmap with Paragon at https://useparagon.com

Welcome to the first episode of the Future Proof Podcast: Building AI Products That Last! In this episode, we chat with Aaron Levie, CEO and founder of Box, about what it takes to build successful Enterprise AI products.We cover:Why good data is a moat / flywheel for AI productsThe importance of access controls in RAGBuilding broad or building deepAPIs, MCPs, A2A(00:00) Clips(00:30) Intro(01:40) Box's AI strategy(04:45) RAG use cases with proven value(09:05) Horizontal vs. deep product strategy(11:14) Prompts as IP (12:05) Importance of secure RAG in enterprise AI(16:28) Ingest vs. agentic queries(19:10) APIs, MCPs and Agent-to-Agent(21:20) Moats and network effects for AI companies(26:49) Risk of bad data(29:48) A few large AI agents vs. many niche AI agentsScale your AI product's integration roadmap with Paragon at https://useparagon.com/?utm_source=podcast/utm_campaign=ep1