Industry leaders share practical lessons for deploying AI agents

Industry leaders told ZDNET that deploying AI agents differs from traditional software launches and shared practical lessons from the trenches on what makes agent projects succeed or fail. Kristin Burnham, writing in MIT Sloan Management Review and reviewing research by Sloan and Boston Consulting Group, said effective agent development requires deliberate choices in control, investment, governance, and design.

She highlighted tensions: constraining agentic systems too much limits effectiveness, while granting too much freedom can introduce unpredictability, and organizations must rethink cost, timing and return on investment and decide whether to retrofit agents into existing workflows or reimagine those workflows.

Governance proved especially important. Nik Kale, who led a team delivering agents that provide expert-level technical guidance to more than 100,000 users, said "confidence isn't accuracy" and that early agents "could respond confidently but incorrectly," requiring heavy investment in retrieval and structured knowledge.

Kale added that "governance can't be retrofitted," urged granting "autonomy in proportion to reversibility," and stressed that "being able to see how a decision was reached matters as much as the decision itself." Practitioners advised starting narrow and prioritizing data quality.

Key Topics

Tech, Ai Agents, Agentops, Nik Kale, Kristin Burnham, Tolga Tarhan