Session Description

As frontier LLMs accelerate code generation across software engineering teams, organizations face a classic realization of Amdahl’s Law. While the code writing phase has been hyper-accelerated, the bottleneck has simply shifted downstream to the sequential, human-in-the-loop code review process. Simply deploying standard AI code review tools or basic chatbot wrappers often exacerbates the issue—resulting in generic feedback, comment fatigue, and a sharp decline in developer engagement over time.

To solve this end-to-end, PhonePe engineered a production-grade, asynchronous, distributed pipeline that orchestrates specialized micro-agents across a directed flow. Operating across a 1,500+ engineer organization and processing hundreds of merge requests daily, this platform has flipped the standard AI engagement curve to achieve a 69% developer actioned rate. This session walks through our battle-tested 5-layer architecture: structural context injection via code graphs, pre-computed staging directories, skill-aware multi-agent delegation, semantic deduplication, and post-merge impact analysis that closes the loop on prompt calibration.

Key Takeaways

• Optimize Context, Don’t Stuff It: Staging pre-computed, file-level diffs and using tree-sitter parsers to calculate blast radius eliminates brittle git command dependencies, slashing agentic loop overhead from 40+ paginations to 2-3 fast file reads.
• Close the Loop with Impact Analytics: True observability requires post-merge workflows that replay suggestions against the final code state to measure real adoption data, turning code review from a fire-and-forget tool into a continuously self-calibrating system.

Target Audience

This session is highly beneficial for:

• Engineering Leaders and DevEx Teams looking to move past AI prototypes and deploy reliable developer infrastructure at scale.
• Platform Engineers focused on building efficient multi-agent loops, minimizing token overhead, and establishing clear observability metrics that prove developer tool ROI.

Speaker Bio

Sanjay Kuniya is an engineer at PhonePe, where they lead the agentic developer-experience platform powering multi-agent code review pipelines, GitLab event orchestration, and review analytics infrastructure. Their work focuses on turning AI prototypes into reliable, production-grade systems that engineering teams actively trust and use.