Own the Outcome, Delegate the Work: A Framework for Testing with AI Agents
SD Times, Tuesday, November 10th, 2026
Speakers at an SD Times event say teams can hand testing work to AI agents but must keep ownership of what quality means.
AI can generate hundreds of tests in minutes but cannot decide which ones matter, Leapwork's Donovan Brady argued at the SD Times Live! AI in Test Supercast on November 5.
Virtual CTO Vaclav Vincalek first questioned the AI push, urging teams to define the task, input, expected output and verification, and warning that more output is not better quality.
Brady said agents can act alone on changes that fit standing policy covering requirements, approved outcomes, mandatory evidence and escalation rules.
Anything touching business outcomes, risk or revenue should be escalated: a moved button is a repair an agent can make, but a changed expected outcome should go back to the requirement owner.