Human Code Review: AI's Next Frontier?

Itamar Friedman, CEO and co-founder of Qodo, delivered a keynote at the AI Engineer World's Fair 2026, titled "The Last Human Code Review: Building Trust in AI-Generated Code." Friedman addressed the critical need for contextual understanding in AI-powered code reviews, arguing that current models, while improving, often lack the specific context of an organization's codebase, tribal knowledge, and architectural standards.

The Two Pillars of Code Review

Friedman began by outlining the fundamental purposes of code review: validation and alignment/learning. Validation ensures code quality, safety, maintainability, and adherence to architectural standards. Alignment and learning, on the other hand, focus on team coherence, skill development, and knowledge transfer. He posed the question of whether human code review remains essential for these two critical functions in the age of AI, and whether the pull request process is still the optimal place for them.

The Context Gap in AI Code Review

Friedman emphasized that the primary bottleneck in achieving trustworthy automated code review is not the AI model's capability but the absence of relevant context. He explained that while AI models can identify bugs, their effectiveness is severely limited without understanding the specific standards, best practices, and historical decisions that guide a particular codebase. "If you don't have the context, even the best model out there will give you different types of bugs and issues," Friedman stated. He highlighted that this context is often fragmented across various documents, communication channels, and, crucially, the uncodified knowledge residing within developers' heads.

Codifying Tribal Knowledge for AI

To bridge this gap, Friedman proposed the need to codify this 'tribal knowledge' and build a 'context lake' or 'context engine.' This system would serve as an interface for both humans and AI agents, allowing them to collaborate effectively. Qodo, as an example, aims to collect and organize team-specific rules and standards, providing this information during reviews. This includes not only explicit rules but also learning from past PRs, discussions, and even production incidents.

From AI to Artificial Wisdom

Friedman argued that the transition from Artificial Intelligence to Artificial Wisdom is crucial for trusting AI-generated code. This involves moving beyond mere detection of errors to a deeper understanding and judgment, mirroring the capabilities of experienced human developers. "If you want to get to a point where judgment is moving to your AI tools, that's AI wisdom where that experience needs to be codified on the right way, the right place for agents and humans," he explained.

The Future of Software Development Governance

The presentation concluded with a vision for the future of software development, characterized by a graph abstraction of the entire development lifecycle. This approach would provide a holistic view of interdependencies and potential conflicts between different code changes. Friedman stressed that to achieve this future, teams must focus on codifying their standards, embracing real-time, self-learning context, and building governance infrastructure that fosters trust and auditability. He predicted that by 2027, organizations that successfully implement these principles could aim for zero outages and zero critical bugs in production.