itssujeeth How mature is your agentic AI practice, really?
I use this model to place organizations on a five-level path from shadow experiments to an agentic enterprise. Pick a level to see who does the work at every stage of the lifecycle, what they actually do, and which tools carry it.
Reading the model
The biggest jump is from level 1 to level 2. Pilots negotiate value, security, and non-functional requirements one project at a time. A platform turns those decisions into defaults, so a low-risk agent ships in days and a high-risk one carries the controls it needs.
The tools named here are illustrative and lean toward AWS and Amazon Bedrock AgentCore. Swap in your own stack. Level 4 is a direction more than a proven state, since very few organizations operate there today.
Four questions that place you fast
- How many agents are running today, and who knows that number?
- Can you show me the eval results for the last agent you promoted?
- Who gets paged if an agent takes a wrong action at 2 a.m.?
- Which agent has a business outcome metric attached to it?
About this model
Authored by Sujith K. Surendran from enterprise agentic AI engagements, 2024 to 2026. It is an explanatory model for CIOs, architects, and platform leads deciding how far to push agents and what has to exist first. It is not a certified assessment instrument, and the tool examples are illustrative. Last updated October 2026.
Two ways to use it
This page shows who does the work, what they do, and which tools carry it at each level and lifecycle stage. The full model at agentic-maturity.itssujeeth.com ↗ adds four placement questions and a reference agent drawn at every level on AWS, Google Cloud, and Azure. Start there to place your organization, then come back here for the roles and tools.