Good ol' SAST to keep your token usage from spiraling (blog.codacy.com)

A new approach to AI-assisted code review emphasizes the effectiveness of deterministic static analysis (SAST) to reduce the costs associated with token usage in AI workflows. Rather than relying solely on stateful large language models (LLMs) that repeatedly process unnecessary context, the proposed method suggests applying deterministic checks first to filter out easily identifiable issues, allowing LLMs to focus on complex judgment calls. This shift in the review process can significantly decrease both human review time and the associated costs, as deterministic analysis consistently identifies known problems without ambiguity. The significance of this approach lies in addressing the escalating costs of code validation in AI development, where the efficiency of traditional human reviews is hampered by an increase in pull requests and complexity. By integrating deterministic static analysis early in the workflow, teams can streamline their processes, enhancing not only cost-effectiveness but also the quality of LLM outputs. This structured approach creates a more efficient AI software development lifecycle (SDLC), ensuring that AI governance remains focused and reduces the repetitive burden on human reviewers, thus fostering a more sustainable coding environment.