Agents think in milliseconds, legacy infrastructure doesn't. LinkedIn, Walmart and Zendesk shared how they closed the gap at VB Transform 2026

The investable signal is not “AI adoption is slowing”; it is that enterprise buyers are moving from model procurement to control planes. That shifts the spend mix toward data lineage, evals, orchestration, security, and observability, which favors platform vendors with distribution across the stack and hurts pure app-layer stories that rely on agent wow-factor alone. In other words, the margin capture migrates from model quality to workflow ownership and infrastructure reuse.

For WMT, the near-term read-through is better opex discipline than revenue upside. Internal agents can compress cycle times and reduce labor drag, but only if governance prevents duplicate tooling and shadow process sprawl; otherwise the company pays an “AI tax” in duplicate systems, audit burden, and cyber risk before it sees savings. The market should care more about whether SG&A leverage improves over the next 2-4 quarters than about any headline productivity narrative today.

Contrarian view: consensus is likely overestimating how quickly agents convert into clean P&L gains and underestimating the integration cost of making them safe. That makes the first winners the boring picks-and-shovels names that sell the plumbing, not the most visible model franchises. The right falsifier is not sentiment; it is whether enterprise software vendors show a faster conversion from AI usage to retained revenue and operating leverage over the next 6-18 months.