Sonar Urges Developers to Treat AI Agents as New Hires

Lauren Hanford, the vice president of product operations at AI code verification firm Sonar, has called for a fundamental shift in how software teams integrate AI agents. Pointing to a Deloitte study predicting that 74% of companies will use agents in some capacity by 2027, Hanford argues that the industry must move beyond treating agent integration as a simple prompting or user interface problem. Instead, organizations need to design robust operating environments that actively guide these autonomous tools.

The push for better agent design comes as developers express significant skepticism about current AI capabilities. According to industry data, 43% of developers worry that AI tools lack sufficient context regarding their specific projects or codebases. While code repositories and API schemas provide some written data, agents consistently miss the unwritten rules, late-night hotfixes, and institutional memory that human engineers share. Hanford notes that because agents cannot learn through osmosis or accumulate experience over time, they remain in a state of being perpetual new hires.

To bridge this gap, practitioners must build explicit scaffolding around their agents. This involves structuring the context window to separate durable rules from task-specific facts, defining clear permission boundaries, and embedding concrete examples of high-quality output directly into the system. Furthermore, developers must pair directional instructions with active intent verification and logging. This ensures that an agent does not simply deliver a statistical average output that compiles but violates internal architectural standards.

For software engineers and IT leaders, this shift changes the metrics of trust. Rather than expecting an AI agent to gradually earn autonomy through cumulative experience, teams must design systems where the agent proves its reliability on every single task. By front-loading acceptance criteria and enabling agents to flag low-confidence decisions before taking action, organizations can prevent invisible technical debt and safely scale their automated workflows.