In my last post, Three Lessons from an Unexpected Sail, I described an afternoon that was supposed to be a relaxing sail to Lopez Pass in the San Juan Islands. Instead, the forecast was wrong, the wind picked up to 20 knots, and two humpback whales greeted us at the entrance to the pass.
Watching whales in the wild reminded me of something I’ve learned from working with AI models. You don’t get the best results by trying to control them — you get better results by understanding how they behave and learning how to work with them.
Here are three habits that have helped me get better results from AI.
1. Don’t try to change the model. Give it better context.
A few years ago, many people assumed every AI application required fine-tuning the model. Today, for most business applications, the opposite is often true. Modern models have much larger context windows and significantly stronger reasoning capabilities. Instead of changing the model, you can usually achieve better results by changing what you provide to it: better prompts, richer business context, relevant documents, examples, and data.
Just as you wouldn’t expect a humpback whale to change its behavior to suit your plans, you shouldn’t expect an AI model to think differently simply because you wish it would. The better approach is to understand how it behaves and provide the context that lets it do its best work.
2. Don’t stop at the answer. Understand the reasoning.
One of the biggest temptations with AI is to accept the first answer as the finished product. That works when the question is simple. As problems become more complex, however, there are usually multiple reasonable approaches. Anyone who uses AI heavily has seen a model completely change its recommendation after being given additional context. The interesting part isn’t that the answer changed. The interesting part is why it changed.
Watching the humpback whales, we weren’t trying to predict every move they would make. Instead, we watched how they behaved, where they surfaced, the direction they were traveling, and how they interacted with each other. Understanding their behavior helped us understand what they were likely to do next.
AI models are much the same. If you only look at the final answer, you miss the thinking that produced it. Understanding the model’s reasoning helps you judge whether the conclusion is trustworthy, what assumptions were made, and if there are alternative approaches worth exploring.
3. Get a second opinion, but compare the reasoning, not just the answer.
When an important decision depends on AI, asking a second model is often worthwhile. You can ask one model to review another’s work, or you can ask multiple models the same question independently. The goal isn’t to vote on which answer wins. The goal is to understand why they disagree.
The same was true with the whales. Seeing one whale surface was exciting. Watching two whales move together revealed much more. You begin to notice patterns—the timing of their surfacing, the direction they travel, and how they respond to one another. Those patterns tell a richer story than a single observation ever could.
Different AI models often make different assumptions, prioritize different evidence, or choose different paths to the same conclusion. Comparing those differences often reveals insights that a single model would never expose.
In my next article, I’ll move from principles to practice. Using concrete examples, I’ll show how prompt engineering, context engineering, and “loop engineering” can help you make better decisions.
Until then, happy whale watching—and happy model using.