Leadership  ·  AI Strategy  ·  For: C-Suite · Board Members · Chief AI Officers

Stop Asking What Someone Does in AI.
Ask What They See.

The surface questions have become noise. What initiatives? What solutions? What impact? Everyone has answers now. The real signal is vision: how someone reads the arc from where AI was to where it is heading. That is what tests genuine depth.

Aug 4, 2026  ·  AI Leadership  ·  Strategy · Vision · C-Suite
3
questions that actually test depth
6yr
arc from narrow capability to reasoning systems
1
shift that separates doers from thinkers in this space

The Question That Has Stopped Working

Walk into any boardroom, any leadership conference, any executive dinner in 2026 and the conversation about AI follows a familiar script. What initiatives are you running? What solutions are you building? What impact are you seeing? How are you deploying it across your organization?

These questions made sense three years ago. They were genuine probes for whether someone had moved past speculation into action. But that window has closed. Today, almost every enterprise leader of any standing has an AI initiative, a pilot deployment, a use case portfolio, and a story about a measurable outcome. The answer to "what are you doing in AI?" has become a press release, not a signal.

The problem is not that the question is wrong. It is that the question has been answered so many times, by so many people, across so many industries, that it no longer differentiates anything. You can prep a convincing answer in an afternoon. And that means you cannot use it to distinguish someone who is genuinely inside this space from someone who has read the right reports and attended the right panels.

The Core Problem

When a question can be answered well by anyone with a good briefing doc, it is no longer a signal of depth. It has become a filter for preparation, not for understanding. The AI conversation needs to evolve past what someone is doing and into how they are thinking.

This matters enormously, because the decisions that will define the next five years in enterprise AI are not operational. They are architectural and philosophical. Where does this go? What gets disrupted that is not obvious today? What are the failure modes that everyone is underweighting? What does the organization need to look like in three years to be on the right side of this? Those are leadership questions, not deployment questions. And you cannot surface them with "what are your current AI initiatives?"

The Arc Is the Test

Here is what actually tests depth: ask someone how they see AI evolving over the last six years and where they think it is heading over the next four. Then stop talking. Listen for whether they can articulate the inflection points, the failed bets, the shifts that genuinely surprised them, and the trajectory they are now tracking.

In 2020, the dominant frame was narrow AI: powerful within a domain, brittle outside it. Models could win at Go, classify images, generate text that sounded plausible, but the underlying assumption was that each capability had to be purpose-built. The idea that a single system could reason across domains, hold context across a conversation, write code, interpret a contract, and plan a multi-step task was treated as a long-range research goal, not an engineering problem to be solved in three years.

Then scale arrived, documented and analyzed through work like Kaplan et al.'s scaling laws (arXiv:2001.08361), and the frame broke. The lesson was not that bigger models were better at the same tasks. It was that scale unlocked capabilities that were not present at all at smaller sizes: reasoning, instruction-following, in-context learning. The capability was not built, it emerged. And that distinction matters deeply for how you think about what comes next.

The right question is not what you have built in AI. It is what you have watched emerge, what you did not predict, and what that taught you about the next inflection.

By 2023, the frame shifted again. Instruction-following models hit the market and the conversation moved from "AI as research" to "AI as product." But most enterprise leaders at that moment made a category error: they treated it as a smarter search engine or a faster content generator. They optimized for the tools they already had rather than rethinking the workflows those tools were sitting inside.

By 2025, the frame shifted a third time. Reasoning models, tool use, and agentic architectures moved AI from responding to a prompt to completing a task. That is not a feature upgrade. It is a qualitative change in what AI is and what it does inside an organization. A system that can call an API, evaluate the result, decide what to do next, and write the output to a database is not a smarter chatbot. It is a new category of worker. And the organizations that understood that early moved decisively. The ones that treated it as a slightly faster version of the thing they already had are now a full cycle behind.

Fig. 1: The Arc of AI Capability, Three Frames in Six Years
Directional illustration. Each frame represents a qualitative shift in what AI systems can do, not merely a performance increment.

Someone who can articulate this arc without prompting, who can name what they got wrong at each transition and what they updated, who can trace the logic from scaling laws to emergent reasoning to agentic systems: that person has been genuinely inside this space. They have not read a summary of it. They have lived it and thought through it. That is depth. And you will never surface it by asking about current initiatives.

The Three Questions That Actually Work

There are three questions that consistently expose the difference between genuine depth and well-prepared familiarity. None of them ask what someone is doing. All of them ask what someone sees.

  1. Where did your model of AI turn out to be wrong, and what did you replace it with? This is the most powerful question because it requires specificity and honesty simultaneously. Someone with genuine depth has been wrong at least twice in this space: once when they underestimated what scale would unlock, and once when they underestimated what reasoning would change about agentic work. The answer to this question reveals whether someone is tracking their own mental models or just tracking the news cycle.
  2. What do you think is being systematically underpriced right now, and why? This tests analytical conviction. Not a safe consensus view. Not what the analyst reports say. What the person actually believes the market and the industry are getting wrong today. Genuine depth produces a specific, arguable answer here. Superficial familiarity produces either a hedge or a talking point.
  3. Where do you think this is in four years, and what is the biggest thing that could make you wrong? This tests vision and epistemic honesty at the same time. A strong answer is not a confident prediction. It is a structured view with named assumptions and named failure modes. The willingness to articulate what would falsify the prediction is itself a signal of how deeply someone has thought about the space.
Why These Work

Each question requires both a position and an accounting of how the position was formed. That cannot be faked with a good briefing. It requires having actually watched AI evolve, updated beliefs in real time, and maintained a running model of where things are heading. That is what separates the people leading in this space from the people following it.

What Genuine Depth Sounds Like

The signals of genuine depth in this space are not credentials or titles. They are a specific kind of reasoning pattern that shows up in how someone talks about AI over time.

First, depth shows up in the willingness to name what failed. Every serious practitioner in this space has a story about an AI initiative that did not work: a retrieval system that hallucinated at scale, an agent that got stuck in a loop, a model that performed brilliantly in evaluation and fell apart in deployment. The person who can tell that story, diagnose what went wrong, and connect it to a structural lesson about how these systems actually behave is inside the space. The person who only talks about successes is not.

Second, depth shows up in the ability to distinguish between the surface of a capability and its mechanism. Being impressed by what a model can do is easy. Understanding why it can do it, what it cannot do even when it appears to, and what that implies for where you would and would not deploy it: that is the analytical layer that most people never reach. It requires reading primary research, not just product announcements. It requires having hit the failure modes personally. And it requires the intellectual discipline to maintain a model of the system rather than just a catalogue of its outputs.

Third, depth shows up in how someone thinks about the next two inflection points rather than the current one. The people who called the reasoning shift early were not prescient. They were tracking the research trajectory and following the logic forward. The same is possible for what comes next: the integration of AI into the physical world through robotics and sensing, the convergence of reasoning and memory into persistent systems that accumulate enterprise knowledge, the shift from AI-assisted decisions to AI-initiated actions. These are not science fiction. They are logical extensions of what is already in the research pipeline. Someone tracking this space closely can articulate that trajectory. Someone who is not cannot.

Fig. 2: Depth Signal Matrix, Surface Familiarity vs. Genuine Immersion
Directional illustration. The gap between familiarity and depth is not about volume of exposure. It is about the quality of the mental model maintained over time.

Why the Conversation Needs to Evolve

The operational questions about AI have served their purpose. They moved organizations from observation to action. That work is not finished, but the limiting factor has changed. The constraint for most enterprise leaders in this space is no longer "are we doing something?" It is "do we see where this is going clearly enough to make the right structural bets now?"

That requires a different conversation. One that spends less time on current initiative portfolios and more time on the underlying theory: how someone reads the relationship between model capability and organizational readiness, where they see the next genuine discontinuity in what AI can do, what they believe about the pace of change in the next cycle compared to the last one, and what that implies for the decisions that need to be made today.

The people who are positioned to lead through the next phase of this are not necessarily the ones with the most deployments. They are the ones who have developed genuine analytical depth about the arc of the technology and the clarity to translate that arc into organizational strategy. That is what the conversation should be testing for. And it requires asking the right questions.

The Standard Going Forward

The question is not what you have built in AI. It is how you see. How you read the trajectory from where AI was to where it is. What you think is happening that others are missing. Where you are placing your bets and why. That is the conversation that tells you whether someone is navigating the next five years or just reacting to them.

There is no shortcut to this kind of depth. It requires sustained engagement with the research, with the deployment realities, and with the failure modes that do not make it into the press releases. It requires intellectual honesty about what you did not see coming. It requires a commitment to updating your model of this space rather than defending the version you built two years ago.

But the leaders who do that work are identifiable. The conversation gives them away immediately. Not because they have more impressive answers to the standard questions, but because they are asking different questions entirely.

Excited about AI, innovation, and growth?

Start a conversation