Who fixes the fix? Accountability for agents in production
Our CTO joined engineering leaders at Adobe to debate how much autonomy remediation agents should get, how every agent-made change gets audited, and what on-call looks like when an agent ships the first fix.
· Bengaluru, India
Utkarsh Ohm
Co-founder and CTO
Overview
Adobe hosted engineering leaders from across Bengaluru for an evening on running AI agents in production. Utkarsh joined the panel to talk about trust: how much autonomy remediation agents should get, and how teams keep every change auditable.
The panel agreed: agents earn autonomy one reviewed fix at a time.
Key takeaways:
Start agents on low-risk fixes and widen their scope as review data builds up.
Treat every agent-made change like a human one: tested, reviewed and logged.
On-call shifts from writing fixes to approving them.
