Verification
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
Verification, in the context of AI coding agents, refers specifically to whether an agent can actually run the thing it has produced — a concept meaningfully distinct from traditional software quality checks like unit tests, linting, or type checking.
Definition and Scope
The speaker in Reflecting on a year of Claude Code draws a sharp distinction between conventional software verification methods and what verification means for agents: "when we talk about verification for agents, it's something slightly different. It's like can the agent run the thing". This frames verification not as static analysis or pre-defined test suites, but as dynamic, executable confirmation — the agent actively running its own output to confirm it works.
Importance
The same speaker emphasizes that "the verification is really important" as a core concern after a year of working with Claude Code, suggesting it is a central bottleneck or quality signal in agentic software development workflows.
Relationship to Traditional QA
By explicitly separating agent verification from unit tests, linting, and type checking, the framing implies that these traditional tools are insufficient on their own for evaluating agent-produced code. The ability to execute the artifact becomes the primary verification criterion, placing runtime feedback at the center of the agent development loop rather than static pre-execution checks.