The Three Categories of Problems
Every problem can be placed on a simple grid based on two questions: do we know the problem exists, and do we know how to solve it?
Category 1: Unknown Problem, Unknown Solution. These are the true blind spots. We don't know the problem exists, so we certainly don't have a solution. Black swan events, undiscovered market needs, and latent systemic risks live here. AI can help with pattern detection, but you fundamentally cannot solve what you haven't conceived of. This category remains largely intractable.
Category 2: Known Problem, Unknown Solution. We see the problem clearly. We feel it. Entire industries complain about it. But nobody has cracked a workable, scalable solution — because the expertise was too scarce, the data too fragmented, or the coordination too expensive. This is the category that AI is blowing wide open.
Category 3: Known Problem, Known Solution. Both the problem and its fix are well understood. Accounting, payroll, standard logistics. These are execution games. AI makes them faster and cheaper, but the opportunity here is incremental, not transformational.
Ten Reasons Category 2 Problems Are About to Become Easy
1. Expertise is no longer a bottleneck.
The defining feature of Category 2 was that you knew what was broken but couldn't find or afford someone who knew how to fix it. LLMs have absorbed vast amounts of specialist knowledge across law, medicine, engineering, finance, and dozens of other fields. The knowledge that used to be locked inside a senior consultant's head is now accessible through a prompt. You still need human judgment for the final mile, but the first 80% of expertise is suddenly free.
2. AI agents can synthesise across domains, which humans couldn't do at scale.
Many Category 2 problems persisted because they sat at the intersection of multiple fields. A rare disease diagnosis requires synthesising genetics, pharmacology, and clinical presentation simultaneously. A complex regulatory question might span tax law, environmental policy, and trade compliance. No single human expert spans all of that. An AI agent can reason across all of it in a single pass. Cross-domain synthesis was the hard part, and it just got easy.
3. The cost of experimentation has collapsed.
When solutions are unknown, you have to experiment — try approaches, test hypotheses, iterate. That used to cost real money: hiring researchers, running pilots, building prototypes. With LLMs and code-generating agents, the cost of generating and testing a candidate solution has dropped from months and lakhs to hours and nearly nothing. You can now explore the solution space a hundred times faster than before.
4. Data exists — we just couldn't process it.
For many Category 2 problems, the raw material for a solution was already out there: buried in research papers, government databases, customer feedback, sensor logs, and industry reports. The bottleneck was never data collection but data comprehension. AI agents can ingest, structure, and extract insight from volumes of information that no human team could get through. The solution was always hiding in the data — we just needed a reader fast enough to find it.
5. LLMs turn tacit knowledge into explicit knowledge.
A huge amount of problem-solving expertise lives as tacit knowledge — the kind that experienced professionals carry in their heads but struggle to articulate or document. LLMs, trained on the written output of millions of such professionals, have effectively distilled a significant portion of this tacit knowledge into a queryable form. Problems that required "finding the right person who just knows" can now be approached by anyone with access to the model.
6. AI agents handle the coordination that made solutions impractical.
Some Category 2 problems weren't unsolved because of a single missing insight but because the solution required orchestrating many moving parts — multiple data sources, multiple stakeholders, multiple sequential decisions. The coordination cost alone made it impractical. AI agents can serve as the orchestration layer: pulling data from different systems, routing decisions to the right people, and managing workflows that would have required an entire project management team.
7. Personalisation at scale is now possible.
Many Category 2 problems were really personalisation problems in disguise. We knew that every student learns differently, every patient has a unique health profile, and every small business has distinct compliance needs. The solution — truly personalised service — was known in concept but impossible to deliver at scale because it required one expert per person. LLMs make one-to-one personalisation economically viable for the first time.
8. The feedback loop between problem and solution has tightened dramatically.
In the old world, you'd identify a Category 2 problem, spend months researching, build a prototype, test it, learn it didn't work, and start over. With AI agents, this loop compresses to days or even hours. You can describe a problem, have an agent generate candidate solutions, evaluate them against real data, refine, and iterate — all within a single working session. The speed of iteration itself changes what's solvable.
9. AI reduces the minimum viable team to near-zero.
Category 2 problems used to require well-funded teams: researchers, domain experts, engineers, coordinators. The minimum viable team to even attempt a solution was expensive, which meant only large companies or well-funded startups could try. Today, a single founder with an AI agent stack can perform work that previously required a team of ten or twenty. This means thousands more people can now attempt to solve Category 2 problems, and the probability that someone cracks each one goes up enormously.
10. The economic incentive is perfectly aligned.
Here's what makes this moment special: Category 2 problems are, by definition, problems that people already know they have and already want solved. There is existing demand with no existing supply. That is the purest form of entrepreneurial opportunity. You don't need to educate the market or create demand — you just need to show up with a solution that works. And because the old cost of solving these problems was enormous (or infinite), even an imperfect AI-powered solution at a fraction of the price represents massive value to the buyer. The margins are extraordinary, the willingness to pay is already there, and the competitive field is still thin.
The Bottom Line
Category 1 problems (unknown problem, unknown solution) remain genuinely hard — you can't solve what you can't see. Category 3 problems (known problem, known solution) are already crowded and commoditised. But Category 2 — where the problem is staring us in the face and the solution was just out of reach — is where the entire game has changed. LLMs and AI agents haven't just made these problems slightly easier. They've made many of them trivially solvable compared to what they were even two years ago. The people who recognise this and act on it are going to build the most valuable companies of the next decade.
