In my conversation with Jentic CTO and Co-Founder Michael Cordner, recorded at AWS re:Invent, he provided a clear and unusually candid look at the practical realities of deploying agentic AI inside the enterprise.
He detailed why APIs built for humans, not machines, are one of the biggest barriers to ROI, and how Jentic helps companies restructure their API environments so agents can operate reliably and safely at scale. Cordner outlined a future of two complementary agent types, highlighted the importance of centralized observability, and emphasized that we are still in the early days of applied AI, where patience and experimentation are essential.
Core Insights
Preparing for AI Agents: Jentic prepares enterprises for agentic AI by restructuring APIs and documentation so they are precise, machine-readable, and ready for large numbers of agents.
Significant Complexity: Many companies underestimate AI complexity, assuming AI can “just figure it out,” when poor documentation and weak infrastructure guarantee poor results.
Operational Excellence: Operational excellence requires centralized authentication, observability, and governance to prevent unmonitored “shadow AI” emerging inside organizations.
How to Achieve ROI: Early ROI comes from agentic systems that accelerate workflow creation, similar to developer tools like Cursor, speeding up experimentation and business logic development.
Key Quotes
Why Enterprises Struggle With AI Adoption
“APIs and documentation up until now have been built primarily for humans and for developers, and for AI to make sense of it, it needs to be a lot more precise and clear. Most organizations just haven’t had enough time to get their APIs and their landscape to the level where AI can be of use. We want to help enterprises be in that 5% that actually sees ROI from AI rollouts.”
The Hidden Risk of Shadow AI
“If there are thousands of agents hitting infrastructure, where are those agents actually living? In a lot of cases that’s shadow IT. People are putting tokens into skunkworks agents with no traceability. You don’t know what systems they’re accessing. Operational excellence starts with centralized, logged, observable agent use.”
The Real Source of ROI in Agentic AI
“The best use of agents is in situations where you determine how to do something, let the agent experiment until it gets it right, and then capture that as a workflow. I think AI is really going to shine by reading process documents, understanding tasks, and creating business logic far faster than humans can. That’s where ROI will come from.”
The Future: Two Classes of Enterprise Agents
“I think in the future there will be two kinds of agents. One type mines processes—trying things until they work and formalizing workflows. The second type takes vague requests, matches them to those workflows, and executes. It’s a simple loop. And yes, it sounds opinionated, but that’s genuinely how I think applied AI is going to play out.”
