Today’s learning didn’t come from a tutorial or a prompt experiment.
It came from a real assignment.

I took on the task of structuring a Facebook group’s travel planning process. Three families, one summer trip, and around twenty potential accommodations spread across Southern Europe. Each option had pros, cons, partial information, opinions, and missing data — the usual chaos of group decision-making.

Instead of letting it stay messy, I treated it like a small AI-assisted project.

Turning conversation into structure

I started by exporting the entire chat conversation into a .md file. That became the backbone.
Then I added information we had discussed outside the chat — context, constraints, preferences.

Only after that did I bring in AI.

I asked ChatGPT to perform deep research on the trip itself: locations, trade-offs, logistics, and relevant comparisons. In parallel, I spun up Claude Code with a very concrete task: publish a page where all collected data lived in one place.

That page wasn’t built after the discussion — it evolved during it.

👉 https://bltg85.github.io/sommarsemester-2026/

As the group conversation continued, the site grew. Options were refined. Notes became structure.

Normalizing the data

Some accommodations quickly became favorites, which meant they accumulated far more data than the rest. That imbalance made comparisons harder.

So I ran another deep research pass — not to find new options, but to normalize the dataset.
Every accommodation was filled with the same categories of information, making trade-offs clearer and discussions calmer.

After that, I did two more targeted research rounds:

• One to identify accommodations we had completely missed
• One to cover neighboring countries we hadn’t seriously considered

Each pass reduced blind spots. Each pass increased confidence.

Parallel AI work, real leverage

Working with ChatGPT and Claude Code in parallel felt… natural.

ChatGPT handled reasoning, exploration, and synthesis.
Claude Code handled structure, publishing, and iteration.

Claude clearly needs to recharge its batteries more often, and it’s obvious that we’ll eventually need to upgrade to a higher tier. But even with current limits, the leverage is undeniable.

What stood out most was how easy it was to collect data in multiple formats and continuously present it in a shared, evolving artifact — instead of scattered messages and opinions.

Still no perfect certainty — and that’s fine

AI didn’t magically answer the hardest question:

Is this the right accommodation? Is this the right vacation?

That uncertainty still exists. And maybe it always will.

But the difference now is that decisions are informed, shared, and grounded in reality — not just gut feeling and incomplete information.

And honestly? With good company, the trip will be great regardless.

What I’m really learning

This wasn’t about travel planning.

It was about learning how to:

• Capture messy human input
• Turn it into structured data
• Run parallel AI workflows
• Build living artifacts instead of static answers

Step by step, I’m getting better at using AI not as a novelty — but as a real tool for navigating complexity.

And that feels like progress.