Spawning parallel AI agents with git subtrees and meta prompts

Calvin Hendryx-Parker joins the podcast to discuss day one deployments, parallel AI workflows, and why Goose excels beyond code.

Staying stuck on your laptop means never solving deployment, secrets, or configuration. In this episode, Calvin Hendryx-Parker, CTO at Six Feet Up and AWS Hero, joins the We Love Open Source podcast to share how day one deployments tackle the whole stack, why git subtrees spawn parallel AI workflows, and how Goose handles business tasks beyond code.

Day one deployment to a sandbox instance solves a whole stack of problems at once: Deployment, secrets, configuration, debugging, triage, logging. Instead of sitting on your laptop where most people stay stuck, deploy to the cloud on day one. This tackles all the side pieces immediately.

Calvin loves Goose, the AI agentic tool from Block. While his team uses Aider and Claude Code, he personally uses Goose not only for code but for building agents to do business tasks: Produce spreadsheets, user stories, customer journeys, then build specifications, epics, requirements, and testing. Goose’s new recipes feature allows making tools that are prompts, very efficient with context engineering. Cost per intelligence is halving every four to six months, making intelligence inexpensive quickly. Smart routing to multiple models simultaneously is coming, getting the best model for the job. ChatGPT-5 already does this under the covers, routing simple stuff to smaller models and bigger stuff to bigger models.

Read more: Meet Goose: The open source AI agent built for developers

The interesting AI use case at Six Feet Up involves meta prompts with git subtrees. Fork off 10 subtrees, one for each GitHub ticket, lay down meta prompts inside each subtree, and spawn sub AI agent processes. Now you’re tending to each, reviewing pull requests, having them lay down tests with a robust set of prompts handling parallel workflows. This is powerful for compliance pieces with lots of YAML or upgrading legacy software to newer versions. AI knows what the end should look like, knows what the beginning looks like, and fills in the blanks quickly.

Senior developers get the most out of AI tools because they have years of knowledge about edge cases. They guide tools with a surgeon’s hand, eliminating boilerplate while catching security backdoors vibe coding folks might miss. But double down on software development life cycle skills: Code reviews, PRs, documentation, tests. Shore these up or you’ll spend 80% of your time triaging unmaintainable code following the Pareto 80/20 rule.

Key takeaways

• Day one deployments solve a whole stack of problems at once: Deploying new projects to sandbox instances on day one tackles deployment, secrets, configuration, debugging, triage, and logging simultaneously instead of staying stuck on your laptop.
• Meta prompts with git subtrees spawn parallel AI agent workflows: Fork off subtrees for each ticket, lay down meta prompts, and spawn sub AI processes that review PRs and write tests. Powerful for compliance YAML and legacy upgrades.
• Goose excels at business tasks beyond code: Use it for spreadsheets, user stories, customer journeys, specifications, epics, and requirements. Cost per intelligence halves every four to six months as smart routing emerges.

Senior developers guide AI tools with a surgeon’s hand while maintaining strong SDLC processes, or you’ll hit the Pareto 80/20 rule spending most time triaging unmaintainable code.

More from We Love Open Source

• Meet Goose: The open source AI agent built for developers
• The AI slop problem threatening open source maintainers
• Stop opening firewall ports and start using identity
• Why 1.3 billion people depend on progress, not perfection
• 5 forces driving DevOps and AI in 2026