Why linting alone can’t govern agentic development

There’s no question that AI agents are changing software development. They can move from an idea to a working implementation quickly, explore alternatives, and take on larger bodies of work that might otherwise have stayed in the backlog. The opportunity is real. The question now is how to turn that speed into reliable, lasting value.

Linting has an important role in that effort. Linters provide fast, local feedback on syntax errors, undefined variables, formatting, and common mistakes. They help teams establish consistent standards without slowing development. In agentic workflows, that immediacy and consistency become even more valuable.

As agents generate larger and more complex changes, teams need a verification workflow that matches the scope of the work. An agent can produce code that is well formatted, idiomatic, and apparently complete. Developers still need confidence that the changes made behave correctly, fit the surrounding system, and meet security and reliability requirements.

“As agents generate larger and more complex changes, teams need a verification workflow that matches the scope of the work.”

This calls for the same engineering discipline applied to any high-leverage system: define the goal, provide useful context and constraints, and verify the outcome before it reaches production.

The original promise of linting has always been immediate, actionable guidance for developers as they write code. That principle should carry forward as teams add the verification layers that agentic development requires.

Verify behavior across the system

Linting excels at inspecting code locally. Many consequential software issues emerge across execution paths, data movement, and interactions between components.

Consider code that looks reasonable in every individual function but passes an unvalidated value through several calls before it reaches a database query, file-system operation, or authentication decision. Or consider a condition that appears defensible…