May Walter is the co-founder and chief technology officer of Hud, a company building function-level runtime intelligence for AI coding agents. Her work gives automated software engineers access to how applications actually behave in production, helping them identify consequential problems, validate fixes, and avoid changes that only look convincing in source code.
Walter previously worked in cybersecurity, vulnerability research, and engineering leadership. She served as CTO of Bond, which was acquired by REEF Technology, and became founding CTO at Santa before co-founding Hud with Roee Adler and Shai Wininger. The company began taking shape in early 2023 and emerged publicly in December 2025 with $21 million in early-stage funding.
Her technical focus developed from operating-system internals and runtime instrumentation into infrastructure for production-aware coding agents. At PyCon Israel 2024, she and Hud researcher Shani Armon explored Python monitoring mechanisms and instrumentation. By 2025, Walter was building a Model Context Protocol server for production runtime context.
Hud’s Runtime Code Sensor maps execution patterns, latency, errors, and external dependencies to individual functions and files, capturing deeper forensic detail when unusual behavior warrants investigation. As Walter explains in her account of Hud’s origins, agents need operational evidence organized around the same code-level structures they use to reason.
- Production context beats better prompts. Source code cannot reveal how often a function executes, which database calls dominate latency, or whether a dependency fails under real traffic. Walter supplies agents with function-level execution evidence before they propose changes, reducing the risk of plausible fixes that worsen heavily used paths.
- Continuous agentic performance optimization. Investigation costs are unpredictable, so engineering teams often postpone performance problems until they become emergencies. Walter’s production-optimization workflow combines GitHub Actions, Claude Code, Model Context Protocol, and Slack to identify N+1 queries, missing indexes, artificial delays, and unnecessarily sequential operations. Opportunities are ranked by business impact, implementation effort, and production risk.
- Verified fixes with human review. Autonomous workflows require stronger safeguards than interactive coding assistance: proposed changes must connect to observed production behavior, pass tests, demonstrate measurable improvement, and clear a human review gate. Walter also treats reviewer attention as a scarce resource, favoring a few high-value recommendations over an avalanche of automated pull requests.
Her longer-term goal is a production-aware software development lifecycle, using runtime evidence to inform predeployment risk assessment, release verification, regression response, and remediation. Hud’s ClickHouse and ClickStack integration extends that approach by connecting broader observability signals with code-level production context.
