Everyone's adopting AI agents. Almost no one can prove the ROI. Here's why

Databricks has published a piece on the gap between AI agent adoption and measurable returns. The core claim: everyone is adopting AI agents, but almost no one can prove the ROI. Amber Roberts, a Staff Technical Marketing Manager at Databricks, is credited with the perspective, described as having spent considerable time on this topic.
Beyond that framing, the report offers little in the way of concrete evidence. No organizations, deployments, benchmark numbers, dates, versions or pricing are named. The headline and framing assert a widespread adoption trend alongside a widespread inability to demonstrate value, but the details behind those assertions — who was surveyed, what was measured, which agents were deployed — are not present.
This suggests the piece is a positioning argument rather than a data-driven study. For teams evaluating agent projects, the practical takeaway is that shipping an agent is not the same as proving it pays off. Value claims typically need baselines, defined success metrics and ongoing evaluation, and that instrumentation is usually an afterthought when teams move fast to adopt a new capability.
Why it matters: product and engineering teams building or buying agents should plan measurement alongside deployment, because adoption without proof of value is hard to defend at budget time. That is my own inference, not a reported finding.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Amber Roberts, Staff Technical Marketing Manager at Databricks, has spent the last...