Rethink Code Reviews

Code review is evolving. As AI-generated code becomes prevalent, so does the complexity of reviewing it: validation (checking functionality and catching edge cases) and understanding (evaluating deeper implications and compliances) are getting more complex, contextual, and individualized.

So how do we build reflective¹, not just reactive, review tools? After spending the past month exploring this space, a few thoughts:

Foundation still matters

Reimagine a simplified version of code review on GitHub

Clarity of structure is more important than ever. Any lack of focus and clear hierarchy will only get amplified with more complex, automated tasks. But this also presents an opportunity: we can, and should, create unified², efficient interfaces make both automated and manual work easier to navigate.

The fluid, dynamic nature

Manging tasks, steering actions, and understanding the system.

The boundary between generating and understanding is dissolving. People might need to frequently switch between asking questions, steering actions, and defining or managing tasks. Should tools specialize by task? Or converge into a single, adaptable surface? Likely somewhere in between.

Balancing social and personal dimensions

What does collaboration look like when automation drives so much personal productivity? Maybe more than ever. We need tools that help users surface problems in complex systems, trace historical decisions, and—crucially—build a shared mental model of how things work.

¹ In Ted Chiang’s story The Truth of Fact, the Truth of Feeling, speech recognition becomes so prevalent that people started losing their ability to write and clearly formulate thoughts. Today, we face a different yet related challenge: content is being generated and regenerated faster than we can process it. Information is everywhere, but understanding is harder to come by.

¹ Many so-called “reinventions” of agent interfaces simply repackage well-established design patterns—lists, comparison tables, kanban boards. These structures are effective, but innovation should be purposeful, not driven by novelty or hype.