25 Sept 2026
Innovator Coffee EP-44 Building the Trust Layer for Enterprise AI
Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. This is Wickey. Follow us to explore the top AI products, ecosystem insights, and the emerging trends.
Podcast Introduction
In this episode of Innovator Coffee, we sit down with Oren Michels, co-founder and CEO of BarnDoor AI, to explore a question that is becoming increasingly important as AI moves from chat to action: what infrastructure do enterprises need before they can truly trust AI agents to do real work?
Oren draws on his experience building Mashery, one of the early API management companies, to explain why he sees striking similarities between the early API era and today’s agentic AI market. The conversation covers category creation, enterprise adoption, fundraising before a market is fully understood, the difference between identity and agent governance, and what separates a durable infrastructure layer from a temporary feature.
Guest Introduction
Oren Michels is the co-founder and CEO of BarnDoor AI, a company building a control plane for agentic AI. BarnDoor focuses on helping enterprises govern and manage AI systems so that agents can safely take actions, connect to tools and data, and generate measurable business value.
Oren previously co-founded Mashery, an early API management company that helped turn APIs from a developer tool into business infrastructure. Mashery was later acquired by Intel. In this conversation, Oren reflects on what he learned from building infrastructure across two very different technology cycles.
Episode Summary
The conversation explores how infrastructure categories are created before customers fully understand that they need them. Oren explains how Mashery emerged from a practical business problem, why its earliest startup-market assumptions turned out to be wrong, and how enterprise demand eventually became the real driver of growth.
The discussion then moves to agentic AI. Oren argues that the central enterprise problem is no longer simply giving users access to AI, but allowing AI to take actions safely and reliably. Traditional identity systems determine what a user or agent is allowed to access, but they do not necessarily govern whether every permitted action is appropriate. That gap, he argues, creates the need for a new agent control plane covering fine-grained access, data protection, model management, spending, and governed automation.
The episode also covers startup fundraising, acquisition lessons, enterprise trust, customer relationships, and why faster software development has not changed one fundamental rule of company building: customers still reward companies that listen, deliver reliably, and solve problems that actually matter.
Chapters / Sub-Summaries
00:00 – 03:30 | From APIs to a New Infrastructure Category
03:31 – 06:26 | Building the First Product and Finding Early Customers
06:27 – 09:37 | Raising Money Before the Market Exists
09:38 – 14:07 | Enterprise Readiness, Trust, and Real AI ROI
14:08 – 18:10 | What an Acquisition Really Teaches You
18:11 – 22:46 | Why Agentic AI Needs a New Control Plane
22:47 – 26:37 | What Makes Infrastructure Durable, and What Hasn’t Changed
26:38 – 31:51 | The Future of AI: From Chatting to Doing
Thanks for Peter Qin helps with the editing
Wickey Wang
*IT Security Compliance Leader & University Faculty (AI security)
*VC advisor and Angel Investor with cybersecurity and AI focus
*GAI Security book co-author
Questions, Suggestions, Feedback and Comments? You can find us in LinkedIn:
https://www.linkedin.com/in/wickey-wang-cisa-six-sigma-green-belt-2aaa913/
