Jensen Huang made a point recently worth sitting with. A software engineer’s purpose, he said, is not to write code. It is to solve problems, and, more importantly, to find problems nobody has yet thought to solve. If AI handles the coding, good. The engineer’s real job was never the syntax. It was the discovery.
This distinction sounds tidy. It is not. Solving a known problem and finding an unknown one are completely different kinds of work. They require different skills, different relationships with the subject, and a different tolerance for uncertainty. Most organisations are built for the first and structurally hostile to the second.
What an unnamed problem looks like
The unknown problem, by definition, has no name. It has no budget line, no brief, no terms of reference. It exists as a gap, a friction nobody has articulated, a consequence the existing frameworks cannot account for.
To make this concrete: in the mid-2000s, hospital-acquired infections were killing tens of thousands of patients a year in the UK and the US. The medical establishment treated this as an inevitable feature of hospital care. Peter Pronovost, a critical care doctor at Johns Hopkins, noticed that a significant proportion of central line infections could be traced to five steps that were routinely skipped during insertion. The steps were known. Nobody had framed their omission as a systemic problem with a systemic solution. He introduced a checklist. Infection rates in his ICU dropped to nearly zero.
The problem was not hidden in any technical sense. Every doctor in every hospital encountered it. But it sat in a gap between what the institution measured and what it did not. Finding it required someone close enough to the work to see the pattern and willing to frame it as a solvable problem rather than an unavoidable cost.
The science parallel
The hallmark of a great scientist has never been the ability to solve hard problems but the ability to find the right ones. Whole fields of inquiry can turn out to be asking the wrong question entirely. The researchers were not incompetent. They were working in the wrong frame.
Ulcer treatment is a useful example. For decades, gastric ulcers were understood as a stress and lifestyle disease. Billions were spent on antacids and dietary interventions. Barry Marshall and Robin Warren’s discovery that most ulcers were caused by a bacterium, Helicobacter pylori, did not require exotic technology. It required someone willing to question a consensus that was so settled it had stopped being examined.
We are seeing something like this now with AI. Entire lines of work are being reframed overnight. In commerce, the assumption has always been that the ground moves slowly enough to plan on. That assumption is breaking down.
Why institutions miss what matters
Unknown problems are invisible to institutions almost by design. The higher you sit in a hierarchy, the more you see what the organisation already knows how to look for. Reporting structures filter information. Incentives shape attention. The unnamed problems stay unnamed, accumulating as unresolved friction, until they arrive as a crisis that surprises everyone except the people closest to the work.
This is not a failure of intelligence. It is a failure of attention. And it is structural. A hospital administrator sees infection rates as a line item. A doctor inserting a central line sees which steps get skipped and why. A retail CEO sees declining footfall as a marketing problem. A shop-floor manager sees that the product range no longer matches what customers are actually asking for. The person who finds the unnamed problem is usually whoever has managed to look at the thing directly, without the filters the institution has placed between itself and reality.
You cannot write a job description for “the person who will notice what we have all missed.” The knowledge that finds unknown problems does not fit into a deliverable. It cannot be contracted for.
The performance of discovery
What fills the space instead is the appearance of discovery. Organisations appear, in technology, in sustainability, in public policy, with the vocabulary of genuine problem-finding but without the substance.
This is visible in the sustainability sector, where consultancies produce materiality assessments that reliably identify the same set of issues their clients are already equipped to address. It is visible in corporate innovation labs that prototype solutions to problems selected for their palatability rather than their importance. It is visible in government commissions that consult widely and recommend narrowly, because the terms of reference were drafted to exclude the conclusions that would be most disruptive.
In each case, the language of discovery becomes decoration on a fixed position. The conclusions were settled before the work began. The process exists to legitimise them.
The people who do this work
The people capable of genuine problem-finding tend to share a few qualities. They are rooted in a specific domain, not skimming across several (although this is not rule by any means). They have stayed long enough to be changed by what they encountered. They notice what does not fit, and they take that discomfort seriously rather than explaining it away.
They are also, typically, difficult for institutions to manage. Their insights do not arrive on schedule. Their conclusions may contradict the organisation’s stated direction. The value of what they find is invisible before it is created, which means it cannot be justified in a business case.
The unknown problem cannot be pitched easily, because the pitch requires a shared vocabulary that does not yet exist. The work of genuine discovery tends not to arrive through formal channels. It grows from trust, from demonstrated usefulness, from a single piece of well-placed insight that makes someone stop and look twice.
What follows from this
Huang is right that purpose and task are different. But recognising the difference is only the beginning. The harder question is whether organisations will actually make room for the kind of work that finds unnamed problems, or whether they will simply adopt the language of discovery while continuing to optimise for known outputs.
AI makes this question urgent. If AI handles the known problem-solving, and it increasingly will, then the value of human work shifts decisively towards problem-finding. But problem-finding cannot be managed the way task completion can. It requires slack in the system, tolerance for uncertainty, and a willingness to act on conclusions that the institution did not ask for and may not welcome.
Most companies will not do this. They will use AI to solve known problems faster and call it transformation. The few that actually create conditions for genuine discovery will find problems their competitors cannot see. That is the real advantage, and it will not come from the technology. It will come from what the organisation does with the time and attention the technology frees up.
Further reading:
↳ The distinction between purpose and task connects directly to the argument in Productivity Is the Wrong Word.
↳ Why institutions are structurally hostile to unnamed problems is the deeper subject of Why Organisations Cannot Do What They Say They Want to Do.
↳ A concrete version of this problem in AI consultancy is developed in The Missing Function.
Garden notes
- Why the frame cannot see itself — why institutions miss unnamed problems structurally
- Separated knowledge — the person closest to the work and the person making decisions are rarely the same
- Proxy capture — institutions optimise for the known and measurable; the unnamed problem is neither
- Productivity Is the Wrong Word — companion essay