· JD Nayak · Agentic AI · 1 min read
From pilot to platform: the five stages of agent proliferation
Agentic AI will not stay a pilot. Here are the five stages from a single advise-only agent to supervised autonomous operations.
Most organisations have a pilot. Fewer have a plan for what happens when the pilot works. I think about proliferation in five stages.
1. Pilot
A single agent on a single use case, advise-only. The goal is learning, not scale.
2. Pattern
A reusable architecture, with a knowledge model, orchestration and human-in-the-loop gates, proven across two or three use cases. This is where reuse starts paying for itself.
3. Platform
A shared agent registry, tool access through MCP, and common evaluation and observability. Teams stop rebuilding plumbing.
4. Fabric
Agents cooperate across process, equipment, MES and supply chain, with policy-driven autonomy.
5. Autonomous operations
Supervised lights-out regions of the factory, with humans managing exceptions and strategy.
Where organisations stall
The usual blocker is between stages one and two. Each pilot is built bespoke, with its own data access and its own idea of safety. The fix is to treat the pattern as the product and the individual use case as an instance of it.
Why open standards matter
MCP, open ontologies and SEMI standards are what let agents from different teams, and different vendors, work together without lock-in. Choose them early, because retrofitting interoperability is the most expensive kind of integration debt.