Claude Code + 15 repos: how a non-engineer answers every customer question | Al Chen
Al Chen (guest), Claire Vo (host)
In this episode of How I AI, featuring Al Chen and Claire Vo, Claude Code + 15 repos: how a non-engineer answers every customer question | Al Chen explores using Claude Code across repos to deliver precise customer support Public documentation and generic AI answers often fail for enterprise customers who need precise, step-by-step explanations tied to how services actually interact in production.
Using Claude Code across repos to deliver precise customer support
Public documentation and generic AI answers often fail for enterprise customers who need precise, step-by-step explanations tied to how services actually interact in production.
By cloning ~15 service repos into one VS Code workspace, Al uses Claude Code to query the living source of truth (main branches) and generate accurate, code-referenced answers faster than pinging engineers.
A lightweight automation script (written by Claude) keeps every repo updated, making it feasible to rely on code even as multiple releases ship daily.
Al augments code context with Confluence and Slack via MCP integrations, including a “customer quirks” page that tailors deployment guidance to each customer’s unique security and infrastructure constraints.
Reactive Slack support is turned into a virtuous loop by summarizing threads into reusable knowledge-base articles (via Pylon), improving future support, internal enablement, and potentially product roadmap inputs.
Key Takeaways
Enterprise customers don’t want “the docs answer”; they want the system walkthrough.
In complex architectures, customers ask how services cascade together (deployment, caching, auth, callbacks), so answers must reflect real interactions—not just product narratives.
Get the full analysis with uListen AI
Open your IDE at the right “scope” to make AI useful across services.
Loading a multi-repo workspace lets Claude traverse dependencies across repos; you can also narrow or widen the directory level depending on the question to manage context bloat.
Get the full analysis with uListen AI
Treat main-branch code as the most current documentation—then keep it current automatically.
Because code changes daily, Al uses a simple Claude-generated script to git-pull main across all repos, avoiding tedious manual pulls and reducing stale answers.
Get the full analysis with uListen AI
Blend code with internal knowledge sources to answer what code alone can’t.
Using Confluence and Slack MCPs, Claude can incorporate deployment runbooks, prior Slack resolutions, and customer-specific constraints to produce responses that match real environments.
Get the full analysis with uListen AI
Maintain a “customer quirks” page to unlock truly tailored guidance.
Capturing items like secret managers, namespace rules, air-gapped constraints, CRD limitations, and encryption requirements allows Claude to produce deployment steps that feel bespoke and build trust.
Get the full analysis with uListen AI
Humans still add value by compressing, validating, and humanizing AI outputs.
Al edits verbosity, removes telltale AI phrasing, and sanity-checks with engineers when uncertain—especially when future refactors exist only in people’s heads or meeting notes.
Get the full analysis with uListen AI
Turn one-off Slack support into compounding assets.
By generating knowledge-base drafts from long Slack threads (Pylon), teams create up-to-date, in-the-weeds articles faster than formal docs PR workflows, reducing repeat questions and enabling training.
Get the full analysis with uListen AI
Notable Quotes
“I can actually pull all of these repos into my VS Code, and I can now use Claude Code to ask our entire codebase questions.”
— Al Chen
“They don't want the docs answer. They want the step-by-step answer of how all these services cascade together.”
— Claire Vo
“The reality is we can now all live in a little bit more chaos because the AI navigates all that information for us across systems.”
— Claire Vo
“I treat it like my entry level analyst… I’m very relentless when it comes to getting the right answer from Claude or from AI.”
— Al Chen
“Throw it into Confluence, throw it into Notion, throw it into Slack, wherever.”
— Al Chen
Questions Answered in This Episode
What does your 16-line “pull all” script actually do (edge cases, branches, auth, failures), and how would you harden it for team-wide use?
Public documentation and generic AI answers often fail for enterprise customers who need precise, step-by-step explanations tied to how services actually interact in production.
Get the full analysis with uListen AI
How do you decide the right workspace scope (single repo vs. multi-repo vs. higher-level directory) before asking Claude Code a question?
By cloning ~15 service repos into one VS Code workspace, Al uses Claude Code to query the living source of truth (main branches) and generate accurate, code-referenced answers faster than pinging engineers.
Get the full analysis with uListen AI
In your DPL custom command, what’s the exact retrieval order and fallback logic between Confluence pages and code search, and how do you prevent stale or conflicting guidance?
A lightweight automation script (written by Claude) keeps every repo updated, making it feasible to rely on code even as multiple releases ship daily.
Get the full analysis with uListen AI
What structure and fields do you include in the “customer quirks” page to make it reliably useful (and not just a messy notes dump)?
Al augments code context with Confluence and Slack via MCP integrations, including a “customer quirks” page that tailors deployment guidance to each customer’s unique security and infrastructure constraints.
Get the full analysis with uListen AI
Where have you seen Claude Code be confidently wrong in this workflow, and what verification steps catch errors fastest?
Reactive Slack support is turned into a virtuous loop by summarizing threads into reusable knowledge-base articles (via Pylon), improving future support, internal enablement, and potentially product roadmap inputs.
Get the full analysis with uListen AI
Transcript Preview
The minute I realized I couldn't really do my job was when I was trying to reference our public documentation and trying to provide an answer. It just still wasn't coming up with a answer that my customers were looking for.
They don't want the docs answer. They want the step-by-step answer of how all these services cascade together.
What I realized is that I can actually pull all of these repos into my VS Code, and I can now use Claude Code to ask our entire codebase questions.
Did you just say, "Claude Code, write me a script that pulls all these?"
Yeah, yeah. I'm opening up the script right now. It's like, what? Sixteen lines. Didn't have to write this. I just said, "Help me figure out a way to pull the latest main branches into my local repos."
The reality is we can now all live in a little bit more chaos because the AI navigates all that information for us across systems, right? So you can be in your code querying Confluence. It will find the information. You have to be less precious about where and how you store the information.
Throw it into Confluence, throw it into Notion, throw it into Slack, wherever. That ends up being context you can provide to Claude when you are trying to ask it a question about a customer or about your codebase.
Let's give Claude Code a little spiff every time it answers a question correctly. You gotta split your quota with Claude Code.
Yeah, it gives you better answers the more bucks you give it or something.
Coin-operated Claude, that's gonna be my new skill. [upbeat music] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, we have an episode all about harnessing your code to make your customer's experience way better. Al Chen, who's on the field engineering team at Galileo, shows us how he uses their 15 repositories and Claude Code to answer every nuanced customer question that comes across his desk and use that to make the entire customer base and his entire team a lot happier. Let's get to it. This episode is brought to you by Orkes, the company behind open source Conductor, which powers complex workflows and process orchestration for modern enterprise apps in agentic workflows. Legacy business process automation tools are breaking down. Siloed low-code platforms, outdated process management systems, and disconnected API management tools weren't built for today's AI-powered world. Orkes changes that. With Orkes Conductor, you get a modern orchestration layer that scales with high reliability and brings humans, AI, and systems together in real time. It's not just about tasks. It's about orchestrating everything: APIs, microservices, data pipelines, human-in-the-loop actions, and even autonomous agents. So build, test, and debug complex workflows with ease, all while maintaining enterprise-grade security, compliance, and observability. Orkes, orchestrate the future of work. Learn more and start building at orkes.io. Al, thanks for joining How I AI. I am really excited about this episode because we've seen a lot about using your code as documentation. You know, we've heard engineers saying, you know, "Docs and code should be in the repo," product managers saying, "Code can now be my documentation for internally facing assets or as I help draft PRDs." But you're gonna show us how you can use code as an asset to create customer-facing things and solve customer-facing problems. So tell me, what, what problem were you facing when you decided, "I'm just gonna clone the repo and fire up Claude Code and solve some of these problems myself"?
Install uListen to search the full transcript and get AI-powered insights
Get Full TranscriptGet more from every podcast
AI summaries, searchable transcripts, and fact-checking. Free forever.
Add to Chrome