In The Hitchhiker’s Guide to the Galaxy, Douglas Adams describes a civilization that decides to build the most powerful computer ever conceived. They call it Deep Thought. It takes 7.5 million years to process the question: “Answer to the Ultimate Question of Life, the Universe, and Everything”. And when it finally gives its answer, the response is: 42.

Nobody knows what it means.

Not because the computer failed. The computer worked perfectly. The problem was that no one had known how to formulate the right question. And without the right question, the most precise answer in the universe is useless.

Douglas Adams

Right now, we are building Deep Thoughts. Only now they don’t take 7.5 million years to respond; they take seconds. Models are bigger, faster, cheaper. Technological civilization is fascinated by the size of the computer, the speed of the response, and the cost per token.

And we still don’t know what question to ask.

Abundance or Apocalypse

We have been immersed in the greatest technological noise of recent decades for two years. Every week, a new model. Every month, a new narrative. Headlines oscillate between the promise of abundance and the labor apocalypse with a cadence that surprises no one anymore. AI is going to create millions of jobs. AI is going to destroy millions of jobs. AI is going to solve climate change. AI is going to concentrate power in the hands of a few. All at once. With equal conviction. With equal lack of evidence.

This noise is not accidental. It is structural. We live in an attention ecosystem where the intensity of the signal matters more than its accuracy. And in that ecosystem, AI is the most valuable asset of the moment—not as a technology, but as a narrative.

The problem is that we are making strategic decisions within that noise. And decisions made within noise tend to reproduce noise.

Construction Was Never the Problem

The dominant narrative in the world of product and technology goes something like this: whoever builds fastest with AI wins. Whoever optimizes their prompts best. Whoever consumes fewer tokens. Whoever automates their development pipeline first. This narrative has a coherent internal logic: if the cost of construction falls, the competitive advantage will lie in speed.

But there is a fundamental error in that logic. It assumes that the bottleneck was always construction.

It wasn’t. It never was.

The bottleneck was always thinking. Understanding what problem we are actually solving. For whom. Why now. What it means for the solution to work. These questions are not answered by speed. They are not answered by a language model. They are not answered by any tool. They are answered by the serious, uncomfortable, and deeply human work of listening, observing, and questioning.

What has changed with AI is not the nature of the problem. What has changed is the cost of construction. And that change, far from making thinking irrelevant, makes it more critical than ever.

A poorly defined problem, fed into an AI, produces a perfectly constructed incorrect answer. At scale. In minutes. This is not progress. It is acceleration in the wrong direction.

The Space Occupied by the Problem

I have spent years working as a designer and strategist in high-complexity environments. And if there is one thing I have learned in that time, it is that the most costly failures do not happen during construction. They happen before. They happen when an organization aligns around the wrong solution with all the conviction in the world. When the problem being solved is the one that was comfortable to solve, not the real problem. When consensus is confused with understanding.

AI does not eliminate that risk. It amplifies it.

And yet, the conversation in our industry continues to revolve around tools, models, and speeds. We continue to replicate, with new technology, the same mental frameworks that produced the same problems before AI existed.

It is time to change that.

Problem-Driven AI

What I am sharing today is the methodology—my methodology, my thinking model put at the service of a new era. I call it Problem-Driven AI. It is not a methodology about AI. It is a methodology about thinking, which ends in AI.

Its premise is simple: value is not in the construction. It is in everything that happens before it.

I will be publishing the methodology, its processes, and artifacts over time, but I want to lay out its main phases and their “whys” here. The methodology is articulated in five phases that form a spiral—not a linear process, but a system that learns with each iteration and becomes more precise with every turn.

Problem Discovery is the first and most important phase. Not a briefing. Not a kickoff meeting. Real research with the people experiencing the problem. Its sole objective is to produce a definition of the problem that is precise, shared, and externally validated. Without this, everything that follows is well-constructed noise.

Solution Alignment is the iterative process of theorizing solutions, presenting them, gathering feedback, and iterating until there is real organizational consensus—not hierarchical, but genuine—on both the problem and the solution. This phase does not end until that consensus exists. A director saying “go ahead” without understanding the solution is not consensus. It is permission. They are different things and produce different results when the first difficulty arises.

Context Engineering is the hinge phase. The systematic translation of the problem and the solution into a structured, precise, and complete context that the AI can process with fidelity. This is, in my opinion, the highest-leverage skill of the AI era. It is not a technical skill. It is a thinking skill. The quality of what comes out is a direct function of the quality of what goes in. Poor context produces poor results regardless of the model you use.

Think of it this way: if Phase 1 produces the diagnosis and Phase 2 produces the prescription, Context Engineering is the process of writing that prescription in a language the kitchen can execute without ambiguity. The chef—the AI—was not in the consultation. It only has what is written. If it is poorly written, what reaches the plate is not what the doctor prescribed.

The context is built with two types of pieces:

• Agents: Specialized roles that process the context in sequence, each responsible for a part of the translation. The Analyst converts the problem and solution into structured requirements. The PM transforms those requirements into a PRD with clear success criteria. The Architect defines how it will be built technically. The Scrum Master breaks everything down into self-contained units of work, Story Files, that a development agent can execute without asking questions.
• Documents: The artifacts that lock in decisions and prevent them from being lost along the way. The PRD defines what is being built and for whom. The Architecture Doc eliminates technical doubts before they appear during construction. The Story Files are the minimum and complete instructions for each task. And the project-context.md is the project’s constitution: the global rules and decisions that no piece can contradict.

AI Build is where the construction happens. And with high-quality context, this phase tends to be the lowest cost and lowest friction of the entire methodology. The job is not to manage the construction but to protect that what is built remains faithful to what was thought. The construction is not the product. The thinking is the product.

Market Iteration is the closing of the cycle and the start of the next. The product in the market generates signals. Those signals are translated into updates for the context. The system learns. The spiral moves forward. Each iteration is more precise than the last because it knows more than the last.

The Most Important Text of My Career

I have subtitled this article “the most important text of my career” not out of grandiosity. I have called it that because it represents the personal and professional moment I am living through. AI inaugurates a new era, one in which everything will change, for better or for worse. Problem-Driven AI gives me back clarity of thought and action; it separates the noise I receive from the signal I seek.

Because if the cost of construction tends toward zero—and everything indicates that is the direction—then the only real differentiator left to us as designers, as strategists, as engineers working at the intersection of human problems and technological solutions is the quality of our thinking.

Not the speed of our tools. The depth of our questions.

The problem was always the problem. AI didn’t change that. It’s just making it more evident who understands it and who doesn’t.