Skip to content

Emerging research direction · Emerging

Self-Verifying Agentic Systems

AI systems that can test, diagnose, and demonstrate when their work is actually complete—and preserve consistent assumptions across durable workflows.

How does an AI system know when its work is actually complete?

An emerging research direction on executable acceptance criteria, automated validation and repair, and consistent assumptions across durable agent workflows.

The problem

AI systems can produce plausible outputs without demonstrating that requirements are covered, tests pass, or assumptions remain consistent across tools and retries.

The approach

Study self-testing, adversarial test generation, invariant checking, requirement coverage, checkpoint semantics, provenance, compatibility, and explicit stopping criteria.

An emerging research direction on executable acceptance criteria, automated validation and repair, and consistent assumptions across durable agent workflows.

Main contributions

  • Research framing and system design
  • Methods, implementation, and empirical evaluation
  • Open research artifacts and scholarly dissemination