My AI and I Built a Debug Mode (Because Things Kept Breaking)

There’s a truth about AI that nobody likes to admit:

It’s brilliant at writing code… and rubbish at seeing whether any of it works.

So after too many frustrating afternoons spent uploading ZIP files, refreshing pages that led nowhere, and wondering what invisible ghost broke the system this time—my AI and I decided to fix the process itself.

We built a Debug Mode.

Why We Needed Debug Mode

AI excels at generating individual steps:

- Want a search form in HTML? Done.

- Need it wired up to Airtable? It’ll generate the code.

- Need a response formatted for display? Boom—copy-paste ready.

But that’s where the magic ends. When you try to actually implement a multi-step system—like a viewer that pulls camera angles and animations from Airtable in response to a natural language query—it breaks down.

Not catastrophically. Just quietly.

The kind of breakdown where nothing loads and you’re not sure if it's your code, your upload, or the cosmos itself.

The AI-Human Debug Partnership

So we devised a better way.

Our Debug Mode follows a simple 5-step loop:

1. State Snapshot: I tell the AI what I tried, what I expected, and what happened.

2. Suspect List: The AI gives me 3–5 likely culprits—ranked and testable.

3. Elimination Loop: I rule out suspects, and we refine.

4. Next Probe: I ask for the lowest-friction way to test the top theory.

5. Capture + Move On: If it’s too hard or still broken, we label it and keep going.

We created three folders to track outcomes:

/debug/

├── Paused/

├── Skipped/

└── Fixed/

This avoids the death spiral of trying the same thing again tomorrow and expecting better results.

What AI Still Can’t Do (Yet)

Even in 2025, my AI still can’t:

- Test the ZIP file it just told me to upload

- Read the HTML page it generated and deployed

- Tell me whether clicking the button actually did anything

That’s the missing glue. The AI can write each brick, but I still have to stack them, check alignment, and spot the cracks.

But here’s the upside: once we realized that, we stopped treating the AI like a magician and started using it like a proper engineering partner.

The Real Power of Debug Mode

Now, we don’t waste time:

- Guessing what went wrong

- Looping on the same broken bit

- Starting over out of sheer frustration

Instead, we pause what’s stuck, move ahead where we can, and leave signposts for Future Us to revisit with a fresh brain (or better tools).

And Yes—We Debugged This Post Too

Of course, the first draft of this post didn’t work.

The tone was off. It wandered. It forgot its own point.

But we ran Debug Mode on it.

And together—my AI and I—we got it working.

Just don’t ask us to deploy it on Netlify.