Everything is buildable now. That's the problem.

My co-founder came back from GCore Connect in Luxembourg with a story that has been sitting with me since.

Misch Strotz, who runs a creative agency and the AI product LetzAI, showed two videos. The first: Luxembourg in the year 963, the founding year - photorealistic, 4K, built in three days for roughly EUR 1,000 in tokens. The second: a commercial for a new LetzAI feature. Not concept sketches. Production grade. The kind of output that, three years ago, needed a crew, a budget and a quarter.

His argument: AI has crossed from toy to production. Whatever the idea is - a film, an app, a software product - if you can describe it, you can now build it.

And in the same room, on the same day, the panel consensus was: not every idea is worth implementing.

Both of these are true. The problem is what happens in the gap between them.

The filter nobody chose, and everybody used

Here is the uncomfortable service implementation cost used to perform: it killed ideas before anyone had to.

A mediocre product idea in 2019 didn't need a diligence process to die. It needed a quote from an agency, a hiring plan, a six-month roadmap - and it quietly went away. Expense was the filter. It was a terrible filter - it killed good ideas along with bad ones, and it favoured whoever had budget over whoever had judgment. But it was a filter, and every organisation's idea pipeline was shaped around its existence.

That filter is gone. Not weakened - gone. When production quality costs three days and a thousand euros, "can we afford to build this?" stops being a question that eliminates anything.

And here is what I see in review after review: nothing has replaced it. Teams still run idea pipelines designed for a world where most ideas self-terminated. Now nothing self-terminates. Everyone is creating something because they can - and the backlog, the portfolio and the roadmap have become places where ideas accumulate instead of places where they compete.

The mistake got more expensive, not less

The intuitive read is that cheap building lowers the cost of being wrong. Build it, see, throw it away. What's the harm?

The harm is that building was never the expensive part - it just used to look like it. The costs that matter come after the build: the launch that consumes your team's attention, the customers who onboard onto something you'll deprecate, the support and maintenance tail, the positioning confusion when your product does eleven things, the opportunity cost of the thing you didn't build instead. None of those got cheaper. Some got more expensive, because production-grade output creates production-grade expectations.

When building costs nothing, the expensive mistake is no longer failing to build. It's building the wrong thing at production quality - and then living with it.

When the build gets cheap enough, being wrong is the only cost left.

I wrote earlier this year that the build got cheap and being wrong didn't. This is the same law, one step further.

The filter has to move into the judgment

So the selection mechanism has to be rebuilt - deliberately, this time, instead of inherited from the finance department. Four tests, in the order I use them:

1. Who pays, and why now?

   Not "who would find this useful." Who moves money, and what changed in their world that makes this the moment? If the honest answer is "nobody yet, but once they see it..." - that is not an answer, that is a hope with a demo attached.

2. What breaks when it's wrong?

   Every product is wrong sometimes. The question is what that costs the person relying on it, and whether you have an answer beyond "it will usually be right." In regulated and B2B contexts this test alone kills a third of the pipeline - correctly.

3. What would you kill to fund it?

   Attention, not money, is the constrained resource now. If the idea is worth building, it is worth naming what it displaces. A team that adds without subtracting isn't prioritising - it's scheduling.

4. What must be true, and how cheaply can we find out?

   The idea's riskiest assumption, stated out loud, with the smallest possible experiment attached.

How not to become the person who kills everything

The failure mode on the other side is real. A filter tuned too tight kills the odd idea that would have mattered, and worse, it teaches the team to stop bringing ideas at all. Two things keep the filter honest:

Cheap builds are probes, not products. This is where Misch's three-day video earns its place in the argument - not as proof that everything should ship, but as proof that test four has never been easier to run. A EUR 1,000 build that answers "will anyone pay?" is the best diligence money can buy. The discipline is in what happens next: a probe has a question attached, a deadline attached and a kill condition attached. If it has none of those, it isn't a probe - it's a product nobody decided to build.

Kill ideas in the open, with reasons. An idea killed by test two ("we have no answer for when it's wrong") can come back when the answer exists. An idea killed silently just breeds resentment and resubmission. The goal is not fewer ideas. It is a pipeline where every idea meets the same four questions, fast, and the ones that pass get real commitment instead of a slot in the jungle.

The decision is the scarce thing now

For decades, the ability to build was the moat. It is not anymore - Misch's videos are what the end of that moat looks like, and they are genuinely impressive. What's scarce now is the willingness to decide: to say this one, not those nine, and to own the reasons.

The build got cheap. The decision didn't. That's not a problem to fix - it's the job description.