← Back to home

When AI Tools Start Feeding Each Other: Our New Dev Workflow Experiment


Lately, we've been running a workflow experiment that feels like a real turning point in how we build.

The setup is simple but powerful:

Codex session workflow overview Click image to enlarge ↗

The Workflow We Tested

We used Codex/CYOU to handle scoped implementation work in a sandbox, with changes tracked cleanly through GitHub. At the same time, in local VS Code (with Copilot), we reviewed, refined, and validated changes side-by-side — often in split screen while PRs were still evolving.

That gave us a fast loop:

It felt less like “tool switching” and more like one connected system.

Parallel Project Experiment

One of the most interesting parts: we even tried this across multiple projects at the same time. In this exact moment, we're updating this blog while also iterating on small internal tools using the same process.

Instead of context-switch chaos, we got a surprisingly smooth rhythm:

What Changed for Us

This experiment made something obvious:

We're no longer just using AI tools.
We're building a cycle where each tool improves the others.

The blog documents the experiments. The tools improve from what we learn. Those improved tools make the next experiment better. And each iteration compounds.

That recursive loop — experiment → tool improvement → better experiment — is exactly what makes this exciting.

Why This Matters

This isn't just about speed. It's about creating a living development system that grows with us:

The most exciting part is that the tooling itself evolves alongside the process.

We're not just shipping features anymore. We're evolving the way we build.

← Back to home