From Blind Spots to Merged PRs: Continuous Agentic Performance Optimization - May Walter, Hud

2026-07-19 · Source: AI Engineer · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, long

Summary

Thundra, a runtime intelligence layer provider, has developed an agentic workflow for continuously optimizing production application performance. This system addresses the "leaky bucket problem" where performance issues are often ignored until critical, by automating the investigation and fix verification process. The workflow, built on vendor-neutral infrastructure (e.g., GitHub Actions, Cloud Code), captures function-level runtime intelligence to identify high-ROI optimization opportunities. It uses a "prod-to-code" approach to provide agents with precise context, enabling them to analyze traces, queries, and latencies, then propose and verify fixes by rerunning tests. Challenges included "plausible unverified" suggestions, complex database queries, and "lazy fixes," which were mitigated by focusing on thorough methodology and robust context. The system generates human-friendly reports for prioritized issues, aiming for high confidence (80-90% trust) before human review, rather than automatically merging numerous pull requests.

Key takeaway

For MLOps Engineers or AI Engineers struggling with reactive performance fixes, consider implementing agentic workflows to proactively identify and verify optimization opportunities. Your team can automate the investigation of production issues at a function level, ensuring fixes are validated before human review. This shifts from crisis mode to continuous improvement, allowing you to focus engineering time on high-ROI changes and build trust in AI-driven code modifications.

Key insights

Automating performance investigation and verified fixes with AI agents can proactively address the "leaky bucket problem."

Principles

• Define what matters with scoring and guardrails.
• Context over cleverness is crucial for agent utility.
• Agentic engineering demands 80-90% trust for autonomy.

Method

The proposed method involves weekly, automated agentic workflows using production context to analyze, score, and flag high-ROI performance opportunities. Agents fix, rerun tests, and verify impact before generating human-friendly reports for review.

In practice

• Implement function-level runtime intelligence.
• Prioritize fixes by business impact and risk.
• Generate human-friendly reports for review.

Topics

• Agentic Workflows
• Performance Optimization
• Runtime Intelligence
• AI in Software Engineering
• MLOps
• Continuous Improvement

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, MLOps Engineer, Software Engineer

Related on AIssential

• 5 Agentic Workflows to Automate Your Data Science Pipeline — Agentic workflows automate repetitive data science tasks, freeing human experts for evaluative decisions.
• Introducing the agent quality loop: AgentCore Optimization now in preview — AgentCore Optimization automates AI agent improvement through data-driven recommendations and rigorous validation.
• Are Your AI Agents Flying Blind? The Truth About AgentOps — AgentOps provides a structured framework for managing, monitoring, and optimizing AI agents in production environments.
• AI agents are entering their rebuild era as enterprises confront the reliability problem — Enterprise AI agent reliability in production hinges on robust workflow orchestration, state management, and recovery, not just LLM performance.
• Agentic DevOps at AWS — AI agents apply reasoning and tool-calling to automate DevOps pain points, enhancing operational velocity and safety.
• Measuring Agents in Production — Production AI agents are basic, human-supervised tools, not fully autonomous systems, delivering tangible value.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

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