Your AI Agent Shouldn't Be Able to Delete the Database — with Camunda's Daniel Meyer
"Just tell the AI not to delete the database" is not a security model — it's a wish. In this IT Visionaries conversation I talk with Daniel Meyer, CTO of Camunda, about why prompts alone can't govern AI agents, and why bolting agents onto legacy processes usually just gives you a faster version of the same broken process.
We get into governance by instruction vs. governance by structure, why so many enterprises can't turn AI spend into real outcomes, and what genuine AI leverage looks like when you reengineer the process instead of decorating it.
What we cover:
- Governance by instruction vs. governance by structure — and why the difference is everything
- Why adding agents to a broken process just makes it fail faster
- What "real AI leverage" looks like — Camunda saw a process run 90x faster
- Applying human-style access controls to AI agents
- Whether enterprise software is going headless
Chapters:
2:16 — Why Business Processes Are Enterprise Algorithms
2:40 — Why AI Investment Isn't Delivering Bigger Outcomes
4:29 — What Real AI Leverage Looks Like
6:35 — Rebuilding the Software Development Process Around AI
11:19 — Why Companies Struggle to Understand Their Own Processes
15:29 — How Harmony Made a Process 90x Faster
16:40 — How ProcessOS Reengineers Business Processes
19:54 — Building AI Agents You Can Actually Trust
21:49 — When an AI Agent Deletes the Database
22:28 — Governance by Instruction vs. Governance by Structure
25:03 — What AI Governance Can Learn From Human Access Controls
38:45 — How to Test and Switch Between AI Models
45:00 — Is Enterprise Software Becoming Headless?
46:51 — The Rise of the Orchestrated Enterprise
Guest: Daniel Meyer, CTO, Camunda — LinkedIn
