Many people understand problems they have never had the means to solve.
They know where a process breaks, what customers keep asking for, or why something that looks sensible on paper fails in practice. But doing something about it requires skills, money or people they cannot easily access. The understanding stays with them. The problem stays too.
AI is loosening that constraint. More people can draw on research, analysis, explanation and technical work beyond their own training. Something that once required assembling a team can become possible to investigate or attempt.
That creates a larger role for intelligence allocation: choosing which problems deserve attention, directing the right combination of human and machine capability towards them, and learning when to change course.
The opportunity reaches well beyond founders and managers. More people can turn what they know about the world into something they can do about it.
From understanding to action
Imagine a community service where residents keep saying they cannot get an appointment. A volunteer who helps people use the service thinks the booking website is the problem. The coordinator agrees and asks for a proposal to replace it.
Previously, that might have meant finding a budget, commissioning advice and waiting. Now the volunteer could use AI to organise feedback, compare approaches and build a simple prototype. A possibility that was out of reach becomes practical to explore.
But the new capability makes the initial question more consequential. “Build a better booking system” sends intelligence towards a particular solution. “Find out why people cannot get an appointment” leaves room to discover that the obstacle is opening hours, transport, confusing instructions or a shortage of staff.
The volunteer's experience helps them notice the problem. It does not establish the cause. The next useful allocation might be a few conversations and some analysis before anyone starts building.
This is where proximity matters. Someone involved in the work can supply the exception, the awkward handover or the recurring misunderstanding that a generic brief leaves out. As I explored in the context essay, the circumstances surrounding intelligence shape what it can usefully do.
There is an analogy with capital allocation. Access to money creates possibilities; deciding what to fund determines what those possibilities become. Access to intelligence creates a similar choice. The return might be a business, a better service, greater understanding or time recovered.
The question is what becomes worth attempting when the person who notices a problem can also help address it.
Fast thinking, slow thinking
Daniel Kahneman's *Thinking, Fast and Slow* offers a useful analogy for how that capability can be organised.
System 1 describes fast, intuitive thinking. System 2 describes slower, effortful deliberation. They are explanatory metaphors for mental processes, rather than separate compartments in the brain.
The expanding range of AI models gives us more ways to match effort to a task. New classification models are designed for quick, focused decisions, such as sorting a request or identifying whether it needs attention. Other models are useful for writing, coding, planning or investigating difficult questions. These capabilities overlap: a language model can respond from familiar patterns or spend longer reasoning and checking evidence.
For the community service, a fast model might group feedback by topic, while a reasoning model helps examine explanations and identify missing evidence. Ordinary software could calculate appointment availability, and a model suited to writing could help turn the findings into a clear proposal.
Several steps could use the same model. With only a few responses, reading them directly may be quickest. The volunteer would check any AI-generated themes against the original comments, including those that do not fit.
When routine steps can be handled quickly and cheaply, deeper reasoning and human attention can be concentrated where they add value. Small tasks that once cost too much to examine individually can become practical to address. The saving depends on the whole process working; mistakes, handovers and corrections still cost time and money.
Choosing the combination is part of the work. It includes recognising where a person should lead.
Sometimes a person is the best option
For some problems, human intelligence may be the most effective choice from the outset.
In the community example, a trusted volunteer may be the best person to speak with residents, including people who have given up trying to book. They may know which question to ask next, recognise hesitation or understand a constraint that never appeared in the written feedback. Staff may know why the current arrangements exist and what changing them would involve.
That is substantive problem-solving. Their judgment and relationships can change the explanation of what is happening and the options worth considering. AI can help prepare questions, organise what is learnt and examine alternatives, while people contribute knowledge that the system would otherwise lack.
The same applies to an experienced practitioner who recognises that a brief rests on the wrong assumption, or someone trusted by both sides of a disagreement who can help them find a way forward.
Good allocation includes choosing human expertise wherever it is likely to produce the best result. It also means making sure that what people discover can influence the decision.
The work must be able to change the plan
Allocation can repeat at every level of an organisation. Leaders judge which problems are most valuable to solve and delegate responsibility for solving them. Managers break those problems into smaller tasks, choose the people or machines each task needs, and bring the results together. The person or AI handling each task can repeat that process, but the pieces still need to add up to a useful result.
For the community service, the coordinator might divide the investigation into residents' experiences, appointment availability and the costs of possible changes. Each piece needs a clear purpose and enough freedom to pursue what turns out to matter.
Suppose conversations and records suggest that many residents cannot attend during weekday working hours. The website may still be awkward, but replacing it would do little for someone who cannot leave work. The volunteer's initial diagnosis needs revising, along with the coordinator's brief.
The person executing the smallest task may discover the largest mistake in the strategy.
That discovery has to travel back up. Staff and residents could help assess whether an evening session is practical, what it would cost and whether it would serve the people currently missing out. The coordinator could then redirect effort from a software project towards a small trial.
Access to intelligence alone does not make this happen. The volunteer also needs a way to challenge the brief, and someone with authority needs to respond. An organisation that gives people powerful tools while refusing to reconsider its priorities can spend less on each task and still waste the whole effort.
AI can participate in this allocation too. It can suggest hypotheses, compare priorities or notice that a result undermines the plan. Keeping responsibility with people leaves room for machines and other people to improve the thinking. This refines the earlier emphasis in this publication on keeping judgment human: substantial parts of the judgment can be shared.
Disciplined curiosity
Following that process takes disciplined curiosity. Curiosity notices a possibility and asks what might be understood or improved. Discipline gives it enough attention to learn something, then uses that learning to decide whether to continue, change direction or stop.
The community group now has a plausible explanation. It still needs to find out whether different hours help. A limited trial might teach it more than another report, provided it checks who uses the new session, whether previously excluded residents can attend, and what demands it places on staff.
If the trial disappoints, the next question changes again. Perhaps transport matters more, or people did not know the new session existed. Intelligence goes towards understanding the result before expanding the intervention.
This extends the learning loop in the memory essay. Outcomes should change what gets commissioned next, including whether the original problem still deserves the same attention.
Experience can improve that allocation. It can help someone recognise which uncertainty matters, which question deserves another hour and when a conversation would be more useful than more analysis. It also needs to leave room for an unfamiliar finding to overturn a familiar explanation.
There is room here for exploration without an immediate practical return. The discipline is in paying attention to what is being learnt. It includes noticing when producing another output has become a way to postpone a decision.
A project can be cheap to start and expensive to own. Good allocation includes knowing when enough has been learnt to act, and when to stop.
Today's allocation shapes tomorrow's judgment
A fool with a tool is still a fool. A more capable tool can help us act on a poor assumption with greater speed and confidence. Access to intelligence does not automatically give us the judgment to direct it.
That judgment has to develop somewhere. The choices we make about delegation affect what we learn as well as what we get done.
Consider a student. AI could complete an assignment, explain a difficult concept, or ask questions that reveal where the student's understanding breaks down. A useful test is what the student can later explain or do without being given the answer. The best allocation for learning may therefore make the student do more thinking.
The same choice exists at work. Someone who delegates every difficult step and sees only finished answers may miss the experiences through which they learn to recognise a weak premise or a convincing mistake. Someone who uses AI to question their reasoning, examine alternatives and test their understanding may develop those abilities.
Kahneman and Gary Klein's work on intuitive expertise helps explain why feedback matters. They argue that reliable intuition depends on an environment with learnable patterns and opportunities to learn through good feedback. Feeling confident is not a reliable test of expertise.
The community volunteer can leave the investigation with both a better service proposal and a better understanding of how to investigate the next problem. That second result depends on staying involved in the questions, evidence and consequences.
An allocation can therefore have two returns: the immediate result and the capability it develops for the future. Education and workplace learning need to take both seriously.
Companies built around intelligence allocation
This also gives a more concrete foundation to the argument in New Zealand's Three-Year Window.
An AI-native company can start with the outcome it wants to deliver, then assemble the models, software and people needed to deliver it reliably. Its structure can follow the problem and change as the available capabilities improve.
Over time, I expect a growing share of the cognitive work in these companies to be performed by AI. In many, most of it may be. Growth could come from directing more machine intelligence towards a proven process, while people contribute where their expertise is most valuable. That changes the relationship between how much work a company can do and how many people it needs to employ.
Those gains could also mean fewer jobs. Whether workers share in the benefits is a separate question from whether the company becomes more productive.
For New Zealand, this gives small teams more scope to build companies serving larger markets. People with a detailed understanding of a problem can attempt solutions that once required a much larger organisation.
Reaching customers and earning their trust still takes work. A company needs customers who benefit from what it does, and a way to deliver that benefit at a sustainable cost. Intelligence allocation provides more ways to organise the effort towards that outcome.
More people able to act
The same possibility extends into ordinary work and life. A teacher can develop something that helps pupils learn. A tradesperson can address a recurring estimating problem. A community group can investigate a failing service and make a practical change.
These are often problems too small to justify a dedicated team, yet large enough to consume meaningful amounts of someone's life. Making them easier to address expands what people can attempt with the experience they already have.
The opportunity will spread unevenly. People still need time, understanding, resources and permission to act. An employee with access to AI may have little influence over which problems get worked on. A community with a persuasive proposal may still lack funding to carry it out.
The promise is a world in which more people can do something about the problems they understand and the possibilities they can see.
I'm Ben Lynch. I write about founders, AI, and what happens next from New Zealand. Say hello at ben@thinkdorepeat.ai.
If you're new here, Start Here is the best place to begin.
If someone comes to mind who understands a problem but has never had the resources to solve it, send this to them.