Give your 'human-level agents' a proper head start with these 3 best practices
Only 19% of businesses have deployed AI agents—here's how to succeed.
Microsoft AI CEO Mustafa Suleyman predicts 'nearly human-level agents' are near, yet a Databricks State of AI Agents report reveals only 19% of organizations have deployed AI agents, mostly limited in scope. Craig Wiley, Databricks' head of AI, identifies three best practices for success: governance, evaluation, and starting small.
Governance involves controlling data access to prevent leakage—e.g., women's health app Flow must ensure its 75 million users never see another's data. Evaluation ensures agent outputs are correct and deterministic, not just suggested by prompts. Starting small means targeting specific tasks rather than entire workflows. Wiley notes CFOs worry about control, value, and cost; these practices directly address those concerns. Clean, organized data underpins all agent work.
- Only 19% of organizations have deployed AI agents, and mostly to a limited extent (Databricks report).
- Governance: enforce deterministic data access rules, as Flow does for 75M users' sensitive data.
- Start small: focus on narrow use cases before scaling to full workflows, per Craig Wiley.
Why It Matters
Enterprises can avoid costly agent failures by following governance, evaluation, and incremental deployment.