Scaling Code Review When AI Writes the Software

On This Podcast

In this conversation, we examine how agentic coding has driven a 150 percent increase in pull request size alongside review times that have nearly doubled, as teams now produce roughly three times more software than before. Jaime describes the risk of automation bias, where developers assume AI-generated code is correct simply because AI produced it, and click to merge without genuine review. We also discuss how this has pushed Codacy to rethink what code review even means when the volume of new code has become almost humanly infeasible to check line by line. Jaime frames this as an industry-wide obsession with maintaining ownership and accountability for software, even as the sheer scale of AI output threatens to erode both.

In this conversation, we examine why Jaime believes deterministic rules remain essential even as AI reviewers become more capable, since AI's non-deterministic nature means it might flag one set of issues today and a different set tomorrow. He describes Codacy's cyborg approach, which pairs a deterministic backbone of static analysis with AI on top to get the best of both worlds. We also discuss how coding harnesses like Claude Code and various IDEs are building in hooks and forced checkpoints precisely because simply hoping an agent will call an MCP tool at the right moment is not reliable enough. Jaime argues this is more than a nice-to-have; it has become a core feature every major agentic tool now needs to bake in.

In this conversation, we examine whether software is losing its value as a defensive business moat now that anyone can prototype an application over a weekend using tools like Lovable or Claude Code. Jaime argues that what actually protects a business is not the code itself but trust and compliance, since real products depend on integrations, standards, and accountability that a vibe-coded mock-up cannot replicate. We also discuss how Codacy positions itself as a builder of trust, helping organisations prove that AI-generated or human-written code meets the quality and security bar needed before it ships to production. Jaime predicts that as the AI hype settles, poor software quality, rather than security breaches alone, will increasingly become the reason people lose their jobs.