How I AI: Zach Davis's 3 Workflows for Enterprise Engineering with AI

On this episode of How I AI, I talked with Zach Davis, the Director of Engineering at LaunchDarkly. He's gone way past just playing around with AI—he's actually weaving it into the day-to-day engineering processes in a large org and code base, affecting everything from documentation to hiring. This isn't just "vibe coding"; Zach is building out solid, scalable AI workflows for a big engineering team. And he was kind enough to share how he's doing it.

Zach walked me through three specific workflows he's put in place at LaunchDarkly that show how AI can help manage even a huge, complicated codebase. We’re going to go through each one step-by-step, with the prompts, code, and insights you need to try them out yourself.

Workflow 1: Centralizing Documentation for Humans and AI Agents

The first thing Zach did was pull all of LaunchDarkly's documentation into one central place inside their repository. This gets around the common headache of having docs scattered across Confluence, Google Docs, and who knows where else. By putting everything in a single docs directory, both engineers and AI agents are looking at the same, up-to-date information. It’s a simple fix for the frustrating problem of finding conflicting or old info.

Step-by-Step Process:

1. Create a Centralized Directory: Zach created a docs directory within his monorepo, organizing content by category (e.g., frontend, backend, accessibility guidelines).
2. Consolidate Existing Documentation: He migrated all relevant documents from various sources into this new directory.
3. Create a Unified Rules System: Instead of separate .cursor rules or .devon files, he created a do_agents_rules directory with universal rules. Individual tool-specific files then simply reference the relevant files from this central location.
4. Leverage AI for Documentation Creation: He used Augment to generate the centralized rule files.

1. Improve Existing Rules: Zach iteratively improved these rules by identifying and addressing areas where AI agents were previously struggling, making the agents more efficient and successful out of the gate.

Example Prompt (Augment):

Create a unified rule file for our AI agents that encapsulates the rules from the cursor rules and the agent's rules files. This file should be easy for both humans and AI agents to understand and use.

Results: This created a single source of truth for all documentation and rules. It led to a big jump in efficiency and made the behavior of their AI agents much more consistent.

Workflow 2: AI-Powered Tech Debt Reduction

With a massive codebase like LaunchDarkly's, technical debt was a real challenge. Zach found a smart way to use AI agents to start chipping away at it systematically.

Step-by-Step Process:

1. Identify Problem Areas: He ran yarn test and piped the output to a log file to identify noisy console logs from tests.
2. Analyze with AI: Zach used Claude to analyze the log file and identify the most problematic areas and their respective severity levels.
3. Prioritize and Create Tasks: This analysis helped generate a prioritized list of tasks in a markdown checklist file in the agents/migrations directory.
4. Assign Tasks to AI Agents: Zach then used Cursor, Devin, and other agents to address these issues individually, one task at a time.

1. Review and Merge Changes: After every task is completed and reviewed, the change is merged into the codebase and the task is marked as complete.

Example Prompt (Claude):

Analyze this test log file and create a prioritized task list for reducing test noise.  Categorize issues by type and severity.  Output in markdown checklist format.

Results: This gave them a clear, repeatable process for tackling tech debt. They were able to reduce noise in tests, resolve accessibility issues, and make it easier for the whole team to contribute to fixing these problems.

Workflow 3: AI-Powered Hiring Process Improvement

Zach even applied this AI-driven thinking to improve the hiring process. He built a custom GPT workflow to help make interview feedback more consistent and useful.

Step-by-Step Process:

1. Create a Rubric: Zach created a detailed rubric for evaluating candidates.
2. Develop a GPT Prompt: He crafted a prompt that included the rubric, examples of excellent and poor scorecards, and instructions for generating feedback and a Slack-ready summary.
3. Analyze Scorecards: Interviewers submit scorecards which are then evaluated by his custom GPT model.
4. Generate Feedback: The model generates feedback highlighting both strengths and areas for improvement. It even produces a concise Slack message for delivering this feedback efficiently.

Example Prompt (GPT):

Evaluate this interview scorecard based on the provided rubric.  Rate the scorecard as Excellent, Good, Fair, or Poor. Provide specific feedback on strengths and areas for improvement.  Also, generate a brief Slack message summarizing this feedback.

Results: The quality and consistency of interview feedback went way up. This helps them make better hiring decisions and gives them a good tool for coaching interviewers.

Conclusion

What I love about Zach's workflows is how practical they are for engineering teams. By centralizing docs, methodically tackling tech debt, and refining the hiring process, he’s showing how to use AI to build solid, scalable processes, not just to find a few shortcuts. His focus on creating workflows that help both the human engineers and the AI agents is a great lesson for anyone trying to bring AI into their company. I really recommend giving these ideas a try.

Sponsor Thanks

This episode is brought to you by:

• WorkOS: Make your app Enterprise Ready today
• Lenny's List: Hands-on AI education curated by Lenny and Claire

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts