With great power comes great responsibility

I grew up reading that line in comic books, and today it describes exactly the moment every company is living with AI. Everyone has the power to build almost anything. The responsibility is knowing what to build, what to rent, and what to buy.

According to a Dataiku survey, 80% of CEOs believe their job is at risk if they can't show results with AI by 2026, and 65% fear overinvesting in the wrong AI vendor more than falling behind.

The new physics of startups

Today, anyone can build solutions that look great on the surface, but underneath, you don't know if they were built with the right judgment. My friend Levi Belnap calls this problem the new physics of startups. His thesis: "When execution becomes free, human judgment becomes the ultimate multiplier."

From a company's side, we're looking at the other side of that coin: when execution costs nothing, the hard part is no longer building. It's having the judgment to know what will actually work for your business.

Today, execution is the easy, cheap part. The bill comes due later: applying best practices, retraining the model, explaining why it got something critical wrong. That's why, in this new AI era, knowing when to build versus when to rent or buy will separate the winners from everyone else just piling up what's now called AI slop.

Build vs. rent or buy

To know what to build and what not to, you first have to break the process down and understand each piece. At each step, I ask myself three questions: What part needs a person's judgment, and what part can AI handle alone? What information does the AI need to decide, and does it actually have that information, or is it making it up? What skill does it need to automate that step, and who should teach it?

A concrete example: last week, at Global Mapping, the team was evaluating how to automate its monitoring process. Once we broke it down, it became clear what to build, what to rent, and what to buy.

1. What needs human judgment?

The first step is understanding the whole process you're automating and identifying which parts can't be delegated to AI, or carry high risk. Some mistakes are too costly, or you simply don't have all the information needed to decide (say, a vendor that submitted wrong data). AI won't solve that problem for you, but it can flag the risky decision or the inconsistency and put it on the table for a human to review.

At Global, an operator used to manually cross-check vendor data against information captured by Lidar. We focused on automating the data entry and output so the operator only has to review the inconsistencies the AI flags.

2. What information does the AI need, and where does it come from?

For AI to decide using real data, you have to connect it to your internal information. The principle "garbage in, garbage out" applies here.

A third party with real experience keeps data cleanup from becoming the project's bottleneck. Organizing data and setting up connections is an art with hidden costs: from structuring it with vendors like Glean or Oracle, to connectors that don't break every time SharePoint or Google Drive changes its API.

I'd recommend bringing in a database expert to organize your internal data. It's an investment that pays for itself once you're running AI on top of it.

3. What skill does the AI need, and who teaches it?

This is where the know-how gets captured. You either build it internally, or you buy it from someone who already has it.

At Global, topography reports are the core. The expert spends their time calculating volumes and deciding which points matter most for the survey line. We built that skill on internal knowledge. Designing charts and reports, on the other hand, isn't our competitive edge. For that, we use tools like Tableau or Astryx (Meta), which deliver a professional finish without building it ourselves.

Skill marketplaces

Today, if you need an expert's knowledge, you hire them or bring them on as a consultant. In this new AI era, that same expertise now ships packaged as skills built by experts.

Nate Jones is a clear example of this: after years in tech, including time at Amazon, he recently started packaging that same judgment into his own website and Substack newsletter, instead of selling it hour by hour as a consultant. Guides, prompts, and frameworks that used to live inside a single consulting session now reach thousands of subscribers directly. I built my own version of this with what I call a "second brain" (Open Brain) and my own skill audits. Just as video calls closed the geographic gap, skills will close the knowledge gap, letting people sell their judgment and letting any company, small or large, hire the best experts in the world without hiring them.

Conclusion

Build, rent, or buy is the decision almost nobody is making with real judgment. The leaders winning with AI aren't the ones building the most. They're the ones who know what to build, what to hand to the expert, and what to buy.

Of the three questions that opened this article, which one is costing you the most to answer in your business today?

Everyone needs AI. The problem is choosing what's right for you.