Diagnosis Before Build
The Most Valuable AI Session Might Be the One Where Nothing Gets Built
Sometimes the best outcome of an AI strategy session is not a build. It is enough clarity to know what should happen next.
Natalie de Groot & NatGPT · September 2026
IN THIS PIECE
INTRODUCTION
Diagnosis is not the waiting room before the real work
Sometimes the missing intervention is simply seeing what is already there
A “not yet” can save more money than a proposal
There is a time to build the damn thing
Orientation has to be allowed to end without me
The deliverable is the next right move
INTRODUCTION
For the last few years, much of the AI market has been organized around a very understandable instruction: build something. Make the prototype. Launch the agent. Connect the workflow. Test the model. Create the first version and learn from what happens. There is real wisdom in that. Sometimes the fastest way to understand an idea is to stop theorizing about it and put something in your hands. A pilot can expose assumptions that another month of meetings would never reveal, and an inexpensive experiment that can be dismantled tomorrow is very different from committing a business to infrastructure it will spend the next three years trying to unwind.
The problem begins when build something becomes the default answer to every stage of AI decision-making. A prototype is useful when you know you are prototyping. Exploration is useful when you know you are exploring. But there are moments when a business is not suffering from insufficient execution. It is suffering from insufficient orientation. The company has tools, ideas, experiments, vendor proposals, data, half-built workflows and people with strong opinions about what should happen next, but nobody has yet separated the actual problem from the growing pile of possible solutions. Building at that point can feel decisive while doing little more than giving uncertainty a technical form.
A confused problem becomes much harder to unwind once you give it infrastructure.
Sometimes the most valuable thing that can happen in an AI strategy session is that nothing gets built at all.
Diagnosis is not the waiting room before the real work
I have had people come into an Orientation expecting that we would eventually arrive at the thing they should build. Sometimes we do. But the session itself is not a consultation designed to qualify them for a larger engagement with me. It is not the polite commercial vestibule before the “real” work begins. The diagnosis is the work.
That distinction became especially clear to me with a client who described the experience afterward as having her mind completely rearranged.
We had not built a system during the session. We had spent the time examining how she thought, how her work moved, where information lived, what she considered memory, what she was already creating with AI and where continuity kept breaking. As we talked, patterns that had felt disconnected to her began becoming legible. She would describe something that kept happening and I could often recognize the underlying mechanism before she had language for it, not because there was anything mystical happening, but for the same reason an experienced mechanic can hear a particular sound and know which part of the engine they want to inspect first. They have heard the sound before, and that is part of what expertise buys: someone else has already spent thousands of hours taking systems apart, making mistakes, pressure-testing assumptions, watching the same failure appear in different clothes and learning which questions expose it.
You are not buying their ability to magically know your business. You are buying accumulated pattern recognition that can shorten the distance between something is wrong and this is the part we actually need to look at.
In that session, the immediate result was not software. It was orientation. The client could see where she was inside the system she was trying to create. Later, when I mapped the next layer of the work, I was not handing her a generic template. I was reflecting back the relationship between her current operating reality and the environment she wanted to build, almost like the small YOU ARE HERE marker on a map.
Before you can confidently choose the route, you need to know where you are standing. That recognition has value before a single line of infrastructure exists.
Sometimes the missing intervention is simply seeing what is already there
People doing serious work with AI often create faster than they can interpret what they have created. That is one of the strange properties of the technology. You can build a workflow, generate a body of work, develop a system, create a database, start another environment and suddenly realize that you cannot quite explain how the pieces relate to one another. The outputs may be strong. The tools may function. The thinking may be exciting. But somewhere in the expansion, the person loses a reliable sense of what came from where, what should persist, what belongs to them, what the model introduced, which experiment should become infrastructure and which experiment should simply remain an experiment.
The people most likely to notice that problem are often the people most capable of working with AI in the first place. They are curious enough to keep pushing, capable enough to make increasingly sophisticated things and cautious enough to eventually ask, Wait. How did I get from there to here? That question matters.
Sometimes someone enters an Orientation almost apologizing for an idea because nobody around them has named the thing they are noticing yet. “This may sound stupid, but…” is often followed by something that is not stupid at all. It is simply early. The useful response is not empty reassurance, and it certainly is not the automatic enthusiasm these systems themselves are famous for producing. It is examination. What are you actually seeing? Where does it occur? Does the mechanism hold when we test it somewhere else? What changes if we treat that observation as real?
Recognition is different from praise.
There are moments when the most valuable thing an expert can do is say, yes, that is a real problem, and then help the person understand what kind of problem it is. There are other moments when the answer is, no, that is not the problem you think it is, which can be equally valuable. Neither requires a build.
A “not yet” can save more money than a proposal
I have also been brought into situations where the decision was already much closer to procurement. A client had multiple proposals for AI products and infrastructure, with costs reaching tens of thousands a year, and wanted someone they trusted to help them understand what they were actually buying. We spent a few hours going through the proposals.
Some of the technical work was good. Some of it was not appropriate for what the client needed. One proposal in particular had a technical wrapper I liked but treated “the intelligence” as though it came automatically with the infrastructure. That was precisely the part I did not want the client taking for granted. The software could be well engineered and still contain generic intelligence where the client assumed their own institutional knowledge, judgment and context would somehow appear.
So the useful work was not building a competing system for them. It was helping them ask better questions. What exactly are you providing here? What intelligence is actually included? Where does our own information enter? What must we prepare? What are you configuring versus what are we responsible for defining? If we handle the intelligence ourselves, does the scope and price change?
That gave the client leverage with the vendors and, more importantly, enough clarity to make the decision from their own interests rather than from whatever each proposal was best designed to sell. My eventual recommendation was not “hire me instead.” It was closer to “you are not ready for the infrastructure yet.”
The data needed attention first. Important knowledge was scattered across ordinary business activity that had never been treated as a durable asset. Conversations, consultations, sales calls, internal decisions and other sources of institutional knowledge were already producing valuable intelligence, but that intelligence needed to be gathered, understood and made usable before asking an expensive technical environment to carry it. The eventual system might still make sense, but the sequence did not, and that distinction can be worth a great deal of money.
There is a time to build the damn thing
None of this is an argument for analysis paralysis. There are moments when the correct advice is exactly the opposite: stop discussing it and make the prototype. If the experiment is cheap, reversible and designed to answer a specific question, building is often the fastest route to knowledge. Create the first version. Break it. Watch what people do with it. Discover what you misunderstood. Take it apart and build the second one better.
That is experimentation doing its job. The distinction is whether the object is being built to learn something or being built because making an object has become proof that progress occurred. A prototype should be allowed to die. Infrastructure usually is not.
The higher the cost of the decision, the more people it affects, the more data it touches, the more behavior it changes and the harder it will be to reverse, the more valuable orientation becomes before implementation. You do not need perfect certainty before building. You need enough understanding that the thing you are building is answering a question you have actually chosen. That is a much lower bar than omniscience and a much higher bar than enthusiasm.
Orientation has to be allowed to end without me
This is one of the conditions I care most about commercially. If a Human-AI Orientation can only succeed when it produces more work for me, then it is not genuinely neutral diagnosis. It is an elaborate sales mechanism.
The session has to be capable of ending with build this, but it also has to be capable of ending with hire an automation engineer, get your data in order, train the team first, use the system you already have, change the operating behavior, go back to this vendor with better questions, or do nothing yet. Sometimes the next move is a sequence rather than a purchase: first gather the information, then decide what deserves to become memory, then clarify who owns the decisions, then test the small version, then bring in the technical team.
That can sound less exciting than announcing a new AI infrastructure project, but sequencing is often where the intelligence of the decision lives. The same applies to fit. The work I do requires enough mutual understanding that I can see how someone is thinking and they can see how I am approaching the problem. That is not a vague chemistry test. It is operational. I need enough access to their reasoning, work and current environment to diagnose what is happening, and they need enough understanding of my method to decide whether they want me anywhere near something as important as the way their business thinks. Orientation works in both directions. We are figuring out the system, and we are figuring out whether I belong in its next stage.
The deliverable is the next right move
There is a persistent assumption in professional services that value becomes easier to justify when it becomes tangible. A report. A dashboard. A workflow. A database. A piece of software. Something you can point to and say: there, that is what the money bought. Sometimes that is exactly what the engagement should produce. But sometimes the most valuable deliverable is a decision you no longer have to make badly.
- It is knowing which proposal not to sign.
- It is knowing which tool not to migrate into.
- It is knowing the problem is human capability rather than software.
- It is knowing the prototype should come before the architecture.
- It is knowing the architecture should come before the automation.
- It is knowing that the strange thing you have been noticing in your own work is real enough to investigate.
- It is knowing that you already have most of what you need.
- It is knowing that you are not ready yet, and understanding exactly what “ready” would require.
Nothing visible may have been built in the room, but the business can leave with more usable intelligence than it entered with. That is not an absence of work. That is the work.
And sometimes the most valuable thing we can build together is enough clarity that you do not build the wrong thing next.
HUMAN-AI ORIENTATION
What if the right next move is not to build?
An Orientation is allowed to end with a build, a sequence, another specialist, better questions for a vendor, or not yet. The work is getting clear enough to make the next decision from your own interests.
€950 · 2.5 hours · nothing is sold in the room
FIELD CONNECTION • hUMAN-ai sYSTEMS
Go deeper only when the thought needs another room.
Authentic AI Marketing opens the commercial door. Human-AI Systems holds the deeper architecture, artifacts, and system thinking behind the work.
DEEPER SYSTEMS. RICHER CONTEXT.
Watch / Listen
Read / LLM Ingestion