Announced. Not built.
The finish line is a press release. That's a strange thing to be racing toward.
Ever been in a room when the company press release hits everyone's inbox at the same time? Phones out, people skimming, and then that slow collective turn, nobody says anything out loud, but you can feel the shared thought settling across the room. That's not really what we've been doing.
To be fair, the sentence usually isn't wrong. It's just not the thing anyone was actually working toward. The announcement had become the outcome.
That happens more than anyone wants to admit. Not because people are careless, but because movement is easy to measure and direction is not. Once something starts moving it creates its own gravity. Before long the work starts organizing itself around what can be said about it rather than what it's supposed to make possible. Speed has a way of disguising direction problems.
There's a version of this that goes differently. When Steve Jobs introduced the iPod, he didn't center what was inside it. He didn't lead with storage specs or engineering achievement. He led with an image.
A thousand songs in your pocket. Not a specification. Not a feature list. An image of your life, slightly better.
The engineers knew exactly what that meant in silicon and circuitry. Everyone else just heard their commute getting better. He chose not to place primary focus on the tool. The meaning came first and the mechanism followed. Generating is cheap now. The work that still requires a human with a point of view is knowing what to generate and why, and being clear enough on the destination that the tool never has to be mentioned.
The output makes the argument. The tool stays invisible.
Fail fast, fail forward. We’ve all heard, my teams have said it. It’s a great line. Hard to actually mean it. What SpaceX built wasn't a philosophy poster. It was a method. Starship blew up. A lot. Each failure was instrumented, analyzed, fed back in. AI didn't make Starship succeed, it made SpaceX fail faster, which is a different and considerably more valuable thing. Nobody watching a booster catch itself on the launch tower was thinking about the software underneath it. They were watching something impossible happen.
A carpenter doesn't tell you about the hammer. He shows you the table.
What's missing isn't effort. It's interrogation.
Something I've always gravitated toward, and where I've found AI genuinely useful, is interrogating the frame rather than accelerating inside it. Not "give me options" but "tell me why I'm wrong." Not "help me move this forward" but "is this actually the thing worth solving?" The question upstream of the work matters more than the work itself because it's what guides everything that follows. Most execution problems are direction problems that got going before anyone stopped to ask.
I also love to cross-examine. Take what one model produces, push it through another with a skeptic's brief, ask where the assumptions live. Because models trained on similar data don't just share capabilities. They share blind spots. Agreement between them isn't confirmation. It might just be two maps with the same roads missing.
"Human in the loop" is meant to reassure everyone that someone responsible is nearby. But it describes a passenger, not a navigator. The more consequential role happens earlier. It's whoever is willing to question where the work is headed before the system gets very good at producing more of it. The one who notices the language getting cleaner while the underlying idea hasn't been challenged. The one who says, without much ceremony, "I'm not sure this is the thing."
That moment rarely lands smoothly. It slows things down. But it's usually where the real work begins.
Before the next press release gets drafted, the question worth sitting with isn't "how are we using AI." That one answers itself quickly and feels like progress without always being it. The harder question is what are you trying to make possible that wasn't possible before, beyond efficiency gains and accuracy. What's the outcome worth announcing when you actually get there.
That question takes longer. It doesn't always produce a clean answer right away. But it has a way of aligning people around something real, something that can be pointed to later without explanation.