Four hours in the air to land where I started

The most productive work session I've had all year happened on a flight that went nowhere.

We took off from Montreal, headed for Atlanta. Two hours in, somewhere over the Carolinas, the captain came on: a wall of storms sitting over the South, no safe way through, and we were turning around. Two hours later we landed back in Montreal, at the exact gate we'd left from. Four hours in the air, zero miles of progress, and the special indignity of clearing customs and security a second time for a country I'd technically never entered. The woman next to me was near tears. I was, I'm a little ashamed to say, having a great time. I had a tray table, a laptop, and a years-old idea that had finally decided to show up.

That boomerang flight was the middle of two days of near-constant travel. And somewhere in those two days, I went from an idea that had never once been written down to a working web product: backend, frontend, tested, with a business plan attached. I built the whole thing in plain English. I directed the work, made every call, and understood what was being built. I just wasn't the one typing the code.

An idea with no paperwork

Here's the strange part about the idea itself: it had lived in the back of my head for years without ever surfacing to conscious thought. Not in a notebook, not in a someday list, not even in the shower-thought rotation. It sat there as a low-grade annoyance I'd never bothered to name. Then, a few days before this trip, it clicked into a sentence: there should be a better way to check cashback portals.

If you've never used one, a cashback portal is a site run by an airline, a bank, or a dedicated cashback company. You click through it before shopping online, buy the thing you were already going to buy, and the portal pays you back a slice of the purchase in cash or airline miles. Same store, same price, same checkout. The only difference is one extra click before you start, and that click is worth real money.

The catch is that rates for the same store vary a lot between portals, and they change constantly. One portal may be paying 2% on purchases at a store like Nike, while a different portal is paying 10% on that same purchase. Next week those numbers flip. A quick check before you buy adds up to serious value over a year of normal shopping. Most people never make that check, because it's tedious.

The gap I'd been living with

There's an incumbent rate-checker in this space, and I want to be clear: it's genuinely good. I've used it for years and it has saved me real money. It aggregates rates across portals and shows you the numbers, reliably, for free.

But it shows you raw rates and stops there. Twelve portals, twelve percentages, figure it out yourself. What I actually wanted every time I opened it was an answer. Is 6% good for this store, or is that the rate it always sits at? Is there a better one right now? Should I wait? The data existed. The judgment didn't. For years I did that judgment manually, in my head, with no memory of what last month's rates looked like.

That was the gap. Not better data. Better answers.

Building at 35,000 feet

So on the first flight, I opened Claude Code and started describing what I wanted. Not code. Sentences. "A shopper should be able to search a store and immediately see whether today's rate is worth acting on." That kind of sentence.

I got lucky on timing. Anthropic had just shipped Fable 5, their new top-end model, days before my trip. Fable is expensive to build with, but it is exceptional at thinking things through, so I used it for exactly that and nothing else. My instruction was almost verbatim: here's what I want to build; plan it out, ask me every question you need answered, then stop. It interrogated me about edge cases I hadn't considered (what happens when a portal quotes its rate in miles instead of cash, and how do you compare those?), and we hammered the plan flat.

Then, with the plan solid, I dropped down to Opus, the cheaper workhorse model, to do the actual building. Plan with the expensive thinker, build with the efficient executor. I didn't invent that split; it's fast becoming the accepted best practice in the published guides on agentic coding with Fable 5. But there's something clarifying about running your team of one the way you'd staff a real project. The architect and the builder don't have to be the same person, or in this case the same model.

The boomerang flight, it turns out, was a gift. Four uninterrupted hours, no Wi-Fi worth the name, no meetings, and nowhere to be, on account of literally heading back to where I started.

What exists now

Two days later, this is real and running: live rate data pulled every day from more than three dozen different portal websites, many of them built on completely different underlying platforms. Over 40,000 stores in the database. Search. A per-store verdict that answers the actual question, which is whether this is a good rate and, if not, where the better one is. User accounts, watchlists, and price alerts for when a store's rate jumps. A newsletter engine, a content system, and admin tooling so I can actually operate the thing. Backend and frontend, built and tested. A business and marketing plan sitting next to the code.

I want to frame this honestly, because the honest version is more interesting than the hype version. This is a stable but rough-edged v1. The branding is placeholder. The aesthetics need a real pass. Some corners of the interface still look like an engineer's first draft, because functionally that's what they are. It works, and it is not yet pretty.

The bottleneck moved

Here's what I keep turning over. Ten years ago, this product would have required either learning to write production code myself or paying a team more than the idea could plausibly return. Five years ago, maybe a technical co-founder and six months. Last week, it required two days of travel and the ability to describe clearly what I wanted.

The bottleneck is no longer "can you build it." For a solo operator with the right tools, the building part is close to solved. The bottleneck now is "do you know what's worth building," and everything that comes after: judgment about the market, taste about the product, discipline about the rough edges. Those are the hard parts, and they were always the hard parts. Plain-English building just collapsed the cost curve enough that a team of one can finally find out.

This product isn't the point of that argument. It's the proof.

Come kick the tires

The site is CheckBeforeYouShop.com. It's live, it's early, and it's rough in the ways I described. If you use cashback portals, or you've been leaving that money on the table because checking felt like homework, try it and tell me what's broken or missing. I read everything, and right now feedback is worth more to me than traffic.

Two days of travel. Four hours in the air to land exactly where I started, customs line and all. And a years-old idea that finally exists, built entirely in plain English. The code was never the hard part.