As AI coding assistants churn out ever greater amounts of code, the first – and arguably most painful – bottleneck that software teams face is code review. A company called Augment Code, which has developed an AI code assistant, yesterday announced a Code Review Agent to relieve that pressure and improve flow in the development life cycle.
The codebases software teams are working with typically are large and messy, and AI models and agents have the fundamental problem of limited insight into the context of that code. According to Guy Gur-Ari, Augment Code co-founder and chief scientist, the company “spent the first year figuring that out. So, given a question or given a piece of code, how do you find the most relevant pieces of code from a repository that might have a million files or more, and how do you do it in a very performant manner?”
Gur-Ari explained that a key differentiator from other code assistants is that the Code Review Agent works at a higher semantic level, making the agent almost a peer to the developer.
“You can talk to it at a very high level. You almost never have to point it to specific files or classes,” he said in an interview with SD Times. “You can talk about, oh, add a button that looks like this in this page, or explain the lifetime of a request through our system, and it will give you good answers, so you can stay at this level and just get better results out of it.”
Augment Code’s early focus with Code Review Agent is on the need for correctness – ensuring the “happy path” works and edge cases are handled. To build developer trust, these review critiques must be highly relevant and avoid generating the noise that causes developers to tune out. This relevance is only achievable when the agent has deep understanding of the code base and is able to review a change within the context of the entire code base, catching cascading effects that a simple line-by-line diff would miss, Gur-Ari said. “When we look at a pull request, we don’t just look at the diff, we look at the context of that diff within the whole code base to see if the change I’m making here, maybe that affects a whole different part of the system negatively. We want to catch things like that.”
Where AI models haven’t been good enough to cover other aspects of the software development life cycle (SDLC) – the so-called ‘outer loop’ of code review, incident triage, fixing CI/CD issues, improving unit testing – today’s agents can, which Gur-Ari said allows Augment Code to expand its coverage of these areas.
This combination of AI writing code and AI reviewing code leads to the question of what role will humans have in a fully automated SDLC? In this emerging model, humans evolve from coders to architects and supervisors. They manage a workflow where different agents handle design, implementation, and testing, but the human is the final check. The future of the SDLC is not about eliminating the developer, but elevating their role to focus on strategic direction, architectural integrity, and the prevention of long-term technical decay.
For now, Gur-Ari said, human intervention is critical. “Imagine you have a process where you have agents doing the design and the implementation and the testing, but at each step of the way you have a developer checking that it’s going in the right direction. I personally don’t think that the models are good enough to remove human supervision,” he said. “I don’t think we’re close to that. One big challenge right now with the agents is that they’re very good at getting to correct code, but they’re pretty bad at making correct design and architecture decisions on their own. And so if you just let them go, they will write correct code but they will accrue a lot of technical debt very quickly. And when you get to 10s of 1000s of lines of code written, if you don’t keep steering them toward correct architecture, you end up with a basically unmaintainable code base.”
According to the company announcement, “expanding into code review is a natural progression — adding the reliability and shared context needed for deeper automation. Augment is building the primitives that let teams shape automation to their unique patterns and architecture. This launch opens up more of those building blocks, with significantly more ahead.”