Forget humans “in” the loop. Harness engineering puts humans “on” the loop.

AI has shifted the focus away from human involvement in the software delivery loop, leading to concerns about the quality and safety of AI-generated code. Thoughtworks' distinguished engineer Kief Morris emphasizes the importance of placing humans "on the loop" to ensure that AI-built software meets production standards, remains secure, and maintains high quality.

Morris warns that relying solely on AI agents can lead to losing track of important details in the code, resulting in a lack of control over the software development process.

Morris advocates for incorporating continuous delivery pipelines within AI-driven workflows to maintain high software quality. He believes that applying the principles of continuous delivery to AI agents can help prevent bugs from slipping into production and enable teams to address issues more efficiently. Instead of focusing solely on how quickly software can be deployed, Morris stresses the need for stronger guardrails to ensure that AI-generated code is safe, secure, and of high quality.

One challenge Morris highlights is the question of what constitutes "good" code, especially in the context of AI-generated software. He urges developers to step back and consider what "good" means for their specific team or organization before relying on AI agents to build code. Morris also warns against delegating too much coding to AI agents, as this can result in cognitive debt – a hidden cost arising from the reliance on AI.

To mitigate cognitive debt, Morris recommends leveraging established software delivery best practices, such as those found in CI/CD pipelines, to enforce quality conventions and maintain control over the software development process.