AI Governance Structures To Ensure Trust - Solo Episode (EP 33)

🎙️ with podcast host Eva Simone Lihotzky
Responsible AI stops being a compliance exercise the moment an agent acts without you. Only 21% of organisations deploying AI agents have a mature governance model for them, and 35% concede they could not shut one down if it went wrong. This episode is for leaders running agentic pilots who could not, if pressed, name the person who answers when one of them fails.
🧭 Episode overview
Eva Simone Lihotzky takes the Harvard Business Review's claim that trust is becoming a competitive advantage and tests it against what she sees inside client organisations. The episode traces the move from generative to agentic to physical AI and argues that each transfer of autonomy changes what trust has to cover: you check an output before it leaves your hands, you check an action once it is already out and handed on, and with physical systems you are trusting a consequence that has already happened. From there it turns to governance, a word she concedes lands badly in most rooms, and to why she thinks the structures behind the technology are the part that compounds. Underneath runs a question she leaves open: technology needs defined inputs, guardrails and outcomes, creative work needs close to the opposite, and she does not claim to know where the line between them sits. It closes on five questions leaders can run against their own systems, and on why a narrative built around efficiency makes resistance the rational response.
🪜 Key themes
Why each step up the autonomy ladder is also a step up the trust ladder, and why the moment you could still check arrives later at every rung
What survives when the models get replaced, and why accountability lines outlast vendor and architecture decisions
Whether governance slows an organisation down, or is the reason it can move at all
Where technical specification and creative freedom have to be negotiated, and why that negotiation never settles
What an unowned decision costs, and the five questions that expose one
🤝 About the host
[PLACEHOLDER - see flag below. Structure to fill: Eva as founder of raidiant, the client work that produces the governance argument, and her governance and assurance background, which is the specific detail that matters for this episode because it is why she can concede how badly the word lands and argue for it anyway. Co-author of '10 Moral Questions: How to Design Tech & AI Responsibly', referenced in the episode as prior art for building governance into the design stage.]
⏱️ Chapter markers
[03:27] Where the human stops being the operator
[08:12] Trusting the output, then the action, then the consequence
[12:16] Outstructuring the competition while the models keep changing
[16:53] What technology needs defined, and what creative work needs left open
[22:14] Five questions, and the decision with no name attached
🔗 Links
Eva Simone Lihotzky on LinkedIn - https://www.linkedin.com/in/evalihotzky/
raidiant - https://www.raidiant.eu
Harvard Business Review, "Responsible AI is Becoming a Growth Strategy" - https://hbr.org/2026/07/responsible-ai-is-becoming-a-growth-strategy
Stanford AI Index 2026 - https://hai.stanford.edu/ai-index/2026-ai-index-report
McKinsey research on agentic AI risk and security concerns: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-era
Informatica CDO Insights 2026: https://www.informatica.com/about-us/news/news-releases/2026/01/20260127-new-global-cdo-report-reveals-data-governance-and-ai-literacy-as-key-accelerators-in-ai-adoption.html
Writer research on AI strategy and rogue agents: https://writer.com/blog/enterprise-ai-adoption-2026/
IBM research on agent decision-making visibility: https://community.ibm.com/community/user/blogs/sarah-bowden/2025/11/18/agentic-ai-is-here-5-key-learnings-from-ibms-lates
10 Moral Questions: How to Design Tech & AI Responsibly: https://www.10moralquestions.com/the-book