Who Owns Quality When AI Makes the Decisions?

Autonomous AI testing agents are already making decisions in your pipelines: what to run, what to retry, what to classify as noise, what to let through to production. The capability is real. The trap is assuming that capability automatically translates into better outcomes.
A capability-without-accountability architecture produces chaos, not quality.
This talk examines what happens when AI agents become first-class participants in the quality chain — and what engineering organizations must build before that happens, to prevent irreversible failures.
The core argument: agentic AI creates an inversion of accountability. Traditional quality traced linearly — developer → QA → operations. Autonomous agents break this. They make hundreds of micro-decisions without individual human approval, decisions that aggregate into a single “green light” no one can fully trace.
For embedded and safety-critical systems, this isn’t a process concern. It’s a safety concern.
The session introduces a practical operating framework: tiered autonomy governance, decision provenance architecture, confidence calibration, and explicit human-agent collaboration interfaces — the structural foundations that separate production-grade agentic testing from compelling demos.
Attendees will leave with a concrete model for transforming “shared responsibility” from ambiguity into a functioning accountability system — before the next 2 AM incident forces the question.