The Proofing Room Issue #12

I dropped a workflow last week on X as a casual response to someone’s question about favorite MCP workflows. Nothing fancy—just showing how I build features while background agents handle the grunt work. The response?

41,000 views. 1,200 bookmarks. 20+ DMs asking "wait, you can DO that?"

Here's the kicker: I've been feeling massively behind. Scrolling through AI Twitter, watching people build autonomous agent swarms, implementing MCPs I hadn't heard of yesterday. Classic imposter syndrome fuel.

But that viral moment revealed something crucial: While I'm worried about keeping up with the bleeding edge, most developers are still copying and pasting between ChatGPT and VS Code.

What feels like table stakes to us is still magic to 99% of builders.

So today, we're pulling back the curtain on the exact workflow that's letting two non-engineers ship production code. Consider this your backstage pass to modern AI development.

Three months ago, my "AI workflow" looked like this:

Last month, we discovered Cursor's background agents. Game changer. Single agents could work on specific tasks without me handholding every step. No more copy-paste gymnastics.

But then we hit a wall.

When you need multiple background agents working in parallel, they operate in complete isolation. No shared context. No coordination. It was like conducting an orchestra where nobody can hear each other.

Enter the plot twist: Claude Code.

Not as a replacement for Cursor, but as the conductor. The orchestration layer that makes everything else sing.

Here's what we've built (and what that viral post barely scratched the surface of):

The Philosophy: Instead of juggling between task trackers, documentation, testing tools, and Slack, create a single source of truth that orchestrates everything else. One brain, multiple hands.

The MCP Stack:

But here's where it gets interesting. We don't interact with these tools directly anymore. We built custom slash commands that handle the complexity:

Each command encapsulates what would normally be 30 minutes of manual coordination. One command, entire workflows.

Remember that feeling when you're building something and you know what you want but not how to implement it? We solved that.

Phase 1: Strategic Research (via Perplexity)

Phase 2: Tactical Documentation (via Context7)

Claude Code orchestrates both phases automatically. It knows when you need strategic guidance versus tactical implementation details. No more context switching between research and coding.

During /project:execute-tasks, something beautiful happens:

But here's the secret sauce: conflict detection. Claude Code analyzes file dependencies before spawning agents. It knows Agent 1 is touching components/ while Agent 2 is in api/. No conflicts, full speed ahead.

When agents might conflict? Sequential execution with context passing. Slower but safer.

Each agent also receives:

The result? What used to take us days now happens in hours. With better quality.

The workflow evolved again this week. Since that viral post, I've been refining how Claude Code fits into our process. Here's our current setup:

For PRD Creation: We use whatever model fits the task—Gemini excels at creative exploration, o3 handles complex logic beautifully, Claude brings nuance to requirements

For Task Execution:

For Quality Assurance: Context-aware automation that knows when to test and when to skip

We're not locked into one tool. We're composing them based on what each does best.

Here's the thing: These tools are constantly evolving. Next week there'll be something new that everyone swears by. Some tools we use might not be "cutting edge." Others we skip might be incredible.

But that's not the point.

The goal isn't to master every new tool. It's to find the combination that lets us focus on what actually matters: building delightful software. The tools should disappear into the background, not become the main event.

Two months ago, David and I were limited in what we could build independently. Now?

David's already a solid frontend developer through his design background. This workflow gives him the tools to build his own ideas without technical blockers.

For me, it means focusing on standardizing these workflows for our entire team instead of getting stuck in implementation details.

And Alan? He can focus on the genuinely complex technical challenges while we handle more of the standard features.

But here's what really matters: We're not special.

This workflow isn't some proprietary secret. It's MCPs, smart tooling, and thoughtful orchestration. The "magic" is knowing how to connect the pieces.

Which brings us to the question...

Look, we're not trying to become another "AI course" grift. But if that viral response taught us anything, it's that people need bridges between "what's possible" and "what's practical."

Tomorrow, I'm planning to record a full walkthrough covering the how, what, and why of this entire workflow. Not just the flashy parts, but the actual implementation details.

We're considering packaging this as:

The question is: Would you actually use this?

Not "sounds interesting" use it. But "I would implement this tomorrow" use it.

Hit reply and tell me:

Virtue Update: Our landing page is live! Check it out and let us know what you think.

Candidly, we made less progress than planned this week. Client work demanded more attention than expected. But that's the reality of bootstrapping. You balance the dream with the revenue.

Workflow Evolution: Every project makes the system smarter. Commands get refined. Patterns emerge. What started as task automation is becoming a development philosophy.

Force Multiplication: This is the real unlock. When the entire team can leverage these workflows, we're not just moving faster. We're expanding what's possible. Each person's capabilities get amplified.

The viral moment was fun, but it raised a real question: How do we scale this knowledge without becoming just another noise source in the AI education space?

So we're asking directly:

Reply and let us know. We read everything, even the brutal honesty.

Next week: Virtue launch prep, momentum building for idea #2, and the results from tomorrow's workflow walkthrough video.

Until then,

Colin Ala& the Torta team