The Same Answer

Four companies that spend billions trying to be different said the same thing in June. Google shipped an agent platform with cryptographic identity per agent. AWS launched a managed knowledge graph so agents know where to find data. Microsoft reframed the bottleneck as context, not capability. Databricks said the visible part of an agent is 1% of the problem.

None of them led with a model announcement. Every one of them led with operations. Runtime, memory, governance, cost tracking, tracing. The parts that have nothing to do with how smart the agent is and everything to do with whether it stays running.

This lines up with what Sinch found last week. Seventy-four percent of enterprise agent deployments got reversed after going live. The vendors saw the same data. They stopped selling intelligence and started selling the infrastructure to govern it.

When competitors converge, it is not a trend. It is the market telling you where the actual problem was. The model was never the bottleneck. The gap was always what happens after you deploy.