How we used AI agents to migrate GitLab rate limiting

2026-07-08 · Source: GitLab · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

GitLab successfully used AI agents to migrate 121 application-level rate-limiting keys from a legacy system to a unified labkit-ruby implementation. This experiment, involving a small team and GitLab Duo Agent Platform, aimed to consolidate Gitlab::ApplicationRateLimiter and a separate Rack-level system. The project shipped 14 specs and over 30 merge requests, rolling out in six cohorts, with Cohort 1 (including pipelines_create, notes_create, user_sign_in) reaching 100% on May 5, 2026. While agents excelled at mechanical tasks like fanning out 95 call sites, human judgment was critical for scope, architecture, rollout, and final review. Challenges included an observability gap that missed a structural collision in shadow mode and a Redis maxclients bottleneck, which halted at 75,000 connections. The experiment demonstrated that AI agents shift bottlenecks from code generation to human review capacity and rollout judgment.

Key takeaway

For MLOps Engineers or AI Engineers migrating legacy systems, integrating AI agents like GitLab Duo Agent Platform can significantly accelerate code implementation. However, you must prioritize robust human-led processes, including adversarial review and meticulous rollout judgment. Ensure your observability distinguishes critical failure modes, and actively audit agent outputs against full inventories. Your "Bob" and "Max" — those who deliberately test and audit — are more crucial than the agents themselves for safe, complete migrations.

Key insights

AI agents accelerate code generation, but human judgment, robust observability, and a strict review loop remain paramount.

Principles

• Observability must distinguish critical failure modes.
• Human judgment is irreplaceable for rollout decisions.
• Agents shift bottlenecks to human review capacity.

Method

Implement a strict loop: spec, adversarial review, implement, verify, adversarial MR review, human review, merge.

In practice

• Use AI agents for mechanical code fan-out tasks.
• Design observability to flag structural data collisions.
• Audit full key inventories against agent-generated changes.

Topics

• AI Agents
• Rate Limiting
• GitLab Duo Agent Platform
• Legacy System Migration
• MLOps
• Observability

Best for: AI Engineer, MLOps Engineer, Software Engineer

Related on AIssential

• GitLab: Built for the agentic engineering era — GitLab's platform is designed to balance agentic speed with enterprise-grade control for the new engineering era.
• Beyond Linting: How AI Code Review Agents Are Learning to Think Like Senior Engineers — AI code review agents are evolving to provide rapid, senior-level architectural and security insights, not just syntax checks.
• Agents Just Passed Humans in Token Usage. And They Burn Far More Than Anyone Budgeted. A Deep Dive With OpenRouter’s COO — Agentic AI usage now dominates, demanding robust inference, tool calling, and cost management.
• Testing AI Agents and Testing With AI Agents Are Two Sides of the Same Coin — The shift to probabilistic AI systems demands a unified quality strategy: both testing with AI agents and rigorously testing AI agents themselves.
• Automation at the speed of Swamp (Friends) — AI agents like Swamp are radically transforming software development by automating complex tasks and shifting focus to architecture.
• GitLab on Google Cloud: Fully managed, compliant, and AI-ready — GitLab and Duo Agent Platform on Google Cloud Marketplace offer managed, compliant, AI-ready solutions with spend retirement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by GitLab.