Zach Davis’s 3 Workflows for Enterprise Engineering with AI

By: Zach Davis Host: Claire Vo Source: How I AI (ChatPRD) Type: podcast

Summary

How I AI episode with Zach Davis, Director of Engineering at LaunchDarkly. Three workflows showing how to weave AI into day-to-day engineering processes in a large org with a massive codebase — beyond “vibe coding” into solid, scalable workflows: (1) Centralizing documentation for humans and AI agents — created a docs directory within the monorepo organized by category (frontend, backend, accessibility), consolidated from scattered Confluence/Google Docs sources, built a unified do_agents_rules directory with universal rules that tool-specific files (.cursor rules, .devon files) simply reference, used Augment to generate the centralized rule files, iteratively improved rules by identifying where AI agents were previously struggling; (2) AI-powered tech debt reduction — ran yarn test piped to a log file to identify noisy console logs, used Claude to analyze the log and categorize issues by type and severity, generated a prioritized markdown checklist in agents/migrations directory, assigned individual tasks to Cursor, Devin, and other agents one at a time, review-and-merge cycle with task completion tracking; (3) AI-powered hiring process improvement — created a detailed interview evaluation rubric, built a custom GPT with the rubric plus examples of excellent and poor scorecards, evaluates interviewer submissions and rates them (Excellent/Good/Fair/Poor) with specific feedback on strengths and areas for improvement, generates concise Slack messages for delivering feedback efficiently.

Key Ideas Extracted

Centralized docs directory as single source of truth: Put all documentation in one monorepo directory organized by category — both engineers and AI agents reference the same up-to-date information instead of scattered Confluence/Google Docs
Unified agent rules with tool-specific references: Create universal rules in a central do_agents_rules directory, then have tool-specific files (.cursor, .devon) reference the relevant central files — avoids maintaining duplicate rule sets
AI-generated rule files with iterative improvement: Use Augment to generate initial rule files, then improve by identifying specific areas where agents were struggling — makes agents more efficient out of the gate
Test log analysis for systematic tech debt reduction: Pipe yarn test output to a log file, use Claude to categorize issues by type/severity, generate a prioritized markdown checklist that agents can work through one task at a time
Agents/migrations directory pattern: Store AI-generated task lists in a dedicated agents/migrations directory within the codebase — gives the tech debt reduction process a clear, trackable home
Custom GPT for interview feedback consistency: Provide a rubric + examples of excellent and poor scorecards in a custom GPT prompt — standardizes evaluation quality and produces Slack-ready feedback summaries
Workflows for both humans and agents: Design processes that help both human engineers and AI agents — centralized docs serve double duty, improving human discoverability while giving agents accurate context
Scalable AI processes over shortcuts: Focus on repeatable, team-wide workflows (documentation, tech debt, hiring) rather than individual productivity hacks — the real value is in building institutional capability

Notes

Published Jul 21, 2025 on How I AI (ChatPRD). ~5 min read. (Note: filename says 2025-09-15 but actual publication date is Jul 21, 2025)
Sponsors: WorkOS, Lenny’s List
Zach Davis background: Director of Engineering at LaunchDarkly
Tools: Augment (rule generation), Claude (log analysis), Cursor (agent tasks), Devin (agent tasks), custom GPT (hiring feedback)
Key pattern: Monorepo docs directory with unified agent rules that tool-specific configs reference
Three companion workflow guides published Jan 8, 2026
Cross-references: Enterprise AI adoption, documentation-for-agents pattern, AI-assisted tech debt, AI in hiring

Raw Content

Re-scraped from ChatPRD 2026-02-16. Full article content captured in Summary and Key Ideas above.