Scale AI Development Without Sacrificing Quality

AI agents can now work on several development tasks at once, which means a lot more code, fast. Whether that code is any good is a separate question.

On the We Love Open Source podcast, Six Feet Up CTO and AWS Hero Calvin Hendryx-Parker talked with host Jason Hibbets about getting more from AI without lowering quality: “Spawning Parallel AI Agents with Git Subtrees and Meta-Prompts.”

Run AI Agents in Parallel

Instead of assigning an AI assistant one task at a time, teams can use Git subtrees and meta-prompts (prompts that direct other AI work) to run multiple AI agents simultaneously.

Calvin described a workflow where developers create separate subtrees for GitHub tickets, give each agent instructions, and let them work in parallel. Engineers then review pull requests and have the agents write tests.

This approach is particularly useful for well-defined tasks like upgrading legacy software or writing compliance configurations. The AI knows what the starting point looks like and what the finished result should be, so it can fill in the gaps quickly.

That frees experienced developers to oversee multiple streams of work rather than execute every task themselves.

Let Experienced Developers Guide the AI

AI can generate code quickly, but it doesn't eliminate the need for engineering judgment.

Experienced developers use AI to cut boilerplate, and they know where things tend to go wrong: edge cases, security backdoors, and the deployment and scaling problems that vibe-coded projects often run into.

They guide the tools with a surgeon's hand, catching problematic patterns early.

Strengthen the Software Development Lifecycle

Weaknesses in engineering processes become harder to ignore as AI accelerates code generation.

Code reviews, pull requests, documentation, and tests are critical. Without them, teams risk generating software faster than they can validate or maintain it, then spending much of the time they saved triaging and debugging code that doesn't work well.

Calvin's recommendation: double down on software development lifecycle (SDLC) practices.

Parallel agents can increase throughput, but that advantage depends on experienced oversight and mature engineering processes.

Watch the Full Interview

Prefer to read? The All Things Open recap covers the discussion, plus day-one deployments and Goose.

Working on AI at your company? See how Six Feet Up can help.