The Question That Outlasts the Product

Most of what we call building is actually imagining, and the imagining is the part that decides whether the building was worth doing. Before a line of code is written, before a prototype ships, before a pitch deck gets polished, someone has to sit with a question long enough to let it change them. The product is just the residue of that inquiry. That is the whole argument, and the reason it matters right now is that the industry has spent three years optimizing the residue while the inquiry got cheaper to skip than it has ever been.

One framework for new product development makes the inquiry explicit by refusing to let you start with the artifact. It asks you to describe the world as it will be once the product exists, and to write that description as a short piece of science fiction before anyone builds anything. The exercise is not creative writing, and treating it as a warm-up misses what it does. It is a forcing function. If you cannot describe the future state clearly enough to make another person feel it, you are not ready to build; you are still guessing, and the guess will be expensive because it will be discovered in code rather than in a paragraph. The story has to contain people doing something they could not do before, which means it cannot hide behind capability language. Nobody writes a compelling scene about a platform that unlocks synergies.

Imagination Is the Scarce Input

This is where most projects actually die. Not in execution, which is now table stakes and getting cheaper by the quarter, but in the gap between what the builder imagined and what the future turned out to want. The uncomfortable part is that this gap is invisible while you are inside it. Execution failures announce themselves through missed deadlines and broken builds, and organizations have decades of accumulated practice at spotting them. Imagination failures look exactly like success right up until launch, because every internal signal, every sprint burndown and every demo, is measuring whether you built the thing correctly rather than whether the thing was worth building.

We do not treat it that way, and our rituals prove it. We celebrate the launch, the funding round, the millionth user. We rarely celebrate the question that survived multiple wrong answers and still held, or the person who kept asking it when switching to a safer, more fashionable query would have been easier and better rewarded. A creative director responsible for the output of six hundred people was once asked to name his own beautiful question, and his answer was not about markets or craft. It was simply: how do I stay inspired? He considers keeping that question alive to be part of the job, not a private matter he handles on his own time. If the person accountable for the creative work of six hundred people has to treat inspiration as a practice rather than a mood, the rest of us have no excuse for assuming it arrives unbidden. Inspiration is not a visitor. It is a habit, and habits are maintained on days when they produce nothing.

Failure Stories Are Calibration

The same discipline applies to the failures, which is where the supply of honest material is largest and the willingness to share it is smallest. Failure stories are the unglamorous infrastructure of any craft that improves over time, and they contain precisely the details that success stories edit out: the wrong assumption nobody challenged, the misread signal, the moment the builder should have pivoted and did not because the original story was still too flattering to abandon. Sharing them is not an act of confession, and framing it that way is what keeps people from doing it. It is an act of calibration. Every failure story is a map of what the future looked like from one particular angle, plus an explanation of why that angle was wrong, and a field with no such maps makes every practitioner rediscover the same terrain alone.

There is a classic dialogue in which a ruler asks a sage to explain the nature of the self, and the sage answers by asking how the ruler came to the meeting. The question shifts the terrain entirely, which is the point. The self is not a thing to be explained; it is a process to be traced. Every version of it is assembled from prior causes, none of which it chose, and none of the versions is permanent. That reframing is worth carrying into products, companies, and every system we build, because we persistently describe them as static objects with fixed properties when they are ongoing processes with dependencies we did not choose and effects we did not anticipate. A platform that succeeds by embedding itself in daily habits will one day find those habits migrated somewhere else, and it will experience the migration as a betrayal rather than as the predictable motion of a process. A payments rail designed around humans clicking buttons will feel the strain when agents start calling APIs instead. Neither outcome is a failure of execution. Both are what happens to systems that never asked the harder question about what they were actually serving.

The harder question is always about dependency, and it decomposes into three that are uncomfortable to ask out loud. What do we rely on that we have never named? What assumptions are baked into the architecture that we have never tested, because testing them would require admitting they are assumptions? Who benefits when this works, and who pays when it does not? A team that can answer those three has a product thesis. A team that cannot has a roadmap, which is a different and much weaker object.

The Moat Moved to the Wiring

The market is currently sending mixed signals about where the answers lie, and the mixture is itself informative. Some companies are racing to embed invisible watermarks in generated text, which quietly shifts the question from whether content is machine-made to who gets to apply the mark. Provenance is being sold as trust, and it does supply some, but provenance is also a power relationship: the entity that controls the mark controls the record, and controlling the record has never been a neutral position for anyone who held it.

At the same time, the cost curves are crossing in a way that inverts the last three years of strategy. Model costs are falling while the cost of wiring systems together climbs. Generating the output is the cheap part now; making it trustworthy, reviewable, and safe to run at scale is the expensive part, and the expense is growing because the surface area of what must be verified grows with every integration. Agents built from smaller models are outperforming larger ones, not because they are individually smarter but because they review each other layer by layer. The intelligence is in the process, not the model. The thing we assumed was valuable, the frontier model, is becoming a commodity, and the thing we neglected, the wiring and the trust layer and the review process, is becoming scarce. Every business plan that assumed the model was the moat is now behind the curve. The moat is who can push the code, who can verify the output, and who can survive the failure modes nobody predicted.

The same pattern shows up in the ordinary business news if you read it structurally. A payments executive built an engine generating seven and a half billion dollars a year by keeping users inside the platform, and her departure is a reminder that systems outlast the people who designed them, which is a compliment to the architecture rather than a diminishment of the architect. The best structures are the ones that do not depend on a single pair of hands, and the test of that only ever runs after the hands are gone. Elsewhere a company shipped a tracking device combining technologies that rarely appear together, and the device is the least interesting part of the story. The real contest is over which locating standard the world decides to adopt, and standards are not won by the best specification. They are won by whichever one other builders choose to embed, which makes the winner a function of developer convenience and installed base rather than of engineering merit.

None of this is new. The form changes constantly and the structure does not move at all. The question that outlasts the product is always the same: what problem are we actually solving, and for whom? The builders who stay in the room long enough to hear the real answer, rather than the first answer that permits work to begin, are the ones whose output survives the hype cycle that carried it in. The rest are writing science fiction about a future they will not be around to see.