If you are trying to map out your AI strategy right now, the signal-to-noise ratio has never been worse. Every day brings a new model, a new benchmark, and a new promise of revolution. However, if we zoom out from the daily hype cycle, we can see that the industry is no longer moving in a single straight line. It is forking.

Part I: The Two Diverging Camps of AI

There are two major trends dominating the world of Artificial Intelligence right now, and they are diverging rapidly.

In one camp, led by the major Large Language Model (LLM) providers, is the mindset that “scale is all you need.” The belief is that by throwing ever more compute at larger and larger datasets, LLMs will continue to make monumental breakthroughs. So far, both the user experience and all of the advanced testing benchmarks show this to be the case.

But in the other camp are the “OG” AI veterans who are predicting that the era of gains via raw scaling is coming to an end. They argue that to drive the next wave of breakthroughs, something else is required. That “something else,” like Voldemort, is rarely named outright.

In this camp are luminaries like Ilya Sutskever (co-founder & former Chief Scientist of OpenAI), Yann LeCun (co-winner of Turing Award with Geoffrey Hinton), and Fei-Fei Li (creator of ImageNet and considered the Godmother of AI): :

“Up until 2020… it was the age of research. Now, from 2020 to 2025, it was the age of scaling… Now, it’s back to the age of research again, just with big computers.” — Ilya Sutskever

“Both supervised learning and reinforcement learning are insufficient to emulate the kind of learning we observe in animals and humans.” — Yann LeCun

“LLMs are too limited and lack a foundation in reality.” — Fei-Fei Li

Like a lot of things in life, both camps are correct. We’re going to see continued advances from existing approaches to LLMs. However, as any savvy veteran will tell you, there is something missing from the current paradigm. I know what it is. And I’m going to tell you what it is and how to prepare your enterprise for it now, Voldemort be damned.

Part II: Understanding the Free Energy Principle (FEP)

To understand what comes next, we have to look at Dr. Karl Friston. He is a famous neuroscientist, the most cited in his field, and a major figure in the world of Artificial Intelligence. His research explores a single, powerful idea: that every living thing—from a tiny cell to a human being—is designed to minimize surprise.

According to his theory, to stay alive, you must constantly predict what is going to happen next. Your brain maintains an internal map of you and your environment to make these predictions. When the world behaves exactly as you expect, you are safe. But when something unexpected happens, there is a gap between your map and reality.

In this model, “Free Energy” is the mathematical term for that gap. It is a measure of your prediction error. When your brain detects this “Free Energy,” it immediately tries to fix it by either updating your internal map (learning) or changing your behavior (acting) to make the surprise go away.

The Squirrel and the FEP

The FEP (Free Energy Principle) states that to maintain its structure and resist disorder (entropy), life plays a survival game where every action is a super-efficient trick an organism uses to avoid surprises.

Imagine a squirrel. As a living organism, it doesn’t like surprises, and it hates wasting energy.

  • Expectation (Prediction): The squirrel expects to find an acorn under the big red leaf.
  • Error (Surprise): The squirrel digs and finds nothing but dirt. This error costs energy because the squirrel’s internal map is wrong.
  • Action (Correction to Fix the Surprise): The squirrel’s brain instantly tells its paws to do the one thing that will fix the surprise fastest while using the least energy: “Move one inch to the right and try again!”
  • Update (Learning): It digs one inch over and finds the acorn! The world now matches its prediction, and the surprise is gone.

According to FEP, the Main Rule of Life is simple: Living things are always trying to minimize surprise.

Part III: The True Value Proposition for Your Business

Why am I confident that the Free Energy Principle is the foundation of the next phase in AI? Because it describes how every successful complex system—from a single cell to a Fortune 500 company—actually functions.

This confirms the core tenet of Organizational Physics: to survive and thrive, a system must constantly ingest new energy and maintain its internal structure against the forces of entropy. FEP provides the mathematical proof for what I’ve been teaching for years, reinforcing the truth that sound mental models always cut across domains.

While FEP is critical for the future of AI—enabling real-time, efficient learning in robots and autonomous agents—its true power is available to you right now. It provides the blueprint for optimizing your company today, regardless of your AI roadmap.

I have stated many times that enterprise AI is an amplifier. It will amplify whatever it touches, be it internal clarity or chaos. The Free Energy Principle clarifies exactly how this works:

  • If your business is designed like a living organism (sensing and minimizing surprise), AI will amplify your efficiency and results.
  • If your business acts like a siloed machine (rigid and disconnected), AI will simply increase your entropy.

Over the coming years, the gap between these two business designs—the adaptive organism versus the rigid machine—will widen tremendously. Here is how to leverage this principle to ensure your business wins.

Part IV: The Business Imperative: Limit Your Free Energy

To be successful, your business must be able to make predictions about what it expects to occur and track what actually occurs. When there’s a surprise, it must quickly close the gap between what has occurred and what was expected. Just like a living system, the goal is to minimize surprise.

Let’s use a business case to see what I mean.

FEP Applied: The Retail Chain

Imagine a national retail chain selling winter coats. The business is the organism, and its survival depends on minimizing “surprise” in the market.

  1. Prediction: The business expects to sell 10,000 blue puffer coats in New England in November. This dictates inventory, staffing, and marketing.
  2. Surprise (The Error): By November 15, actual sales show only 4,000 blue coats sold. This 6,000-unit difference is a massive Prediction Error.
    • If the business is a “siloed machine,” it waits too long to react, wasting capital on 7,000 unsold coats (a high “metabolic” energy cost).
  3. Action to Minimize Error (Cost-Effective Correction): An FEP-aligned business acts immediately. It selects the highest-leverage (i.e., energy-efficient and high-outcome) action to align its reality with its goal: launching a localized 20% flash sale and adjusting the shipping manifest to move coats to a different, high-demand region.
  4. Error Reduction: The gap is closed, the Prediction Error is minimized, and the business secures its revenue goal with minimal excess energy expenditure.

Now, how will enterprise AI amplify these results?

How Enterprise AI Can Limit Surprise to Amplify Results

If the retail chain is designed well (like a “living organism”), Enterprise AI delivers greater success by making the process of minimizing Prediction Error nearly instantaneous and hyper-optimized.

By making predictions smarter, error detection faster, and corrective actions hyper-optimized, Enterprise AI ensures that the retail chain operates as the ultimate self-regulating organism.

However, I must call out a critical nuance from Organizational Physics: Optimizing for the known, short-range environment does not guarantee long-range success.

  • Don’t let efficiency overpower effectiveness. A living system must retain the “slack” and capability to continuously try new things. If you use AI to optimize away all variance, you lose the ability to adapt to changing conditions.
  • Don’t let short-range pressure overpower long-range needs. There is a reason the squirrel harvests and stores nuts for the winter. While your AI agents are optimizing for this month’s sales, you must ensure you are still designing and planning for the long-range future.

Part V: Design Your Business Like a Living System

Your core takeaway from this article should be that every living system tries to limit its surprises. It does this not by being overly controlled or prescriptive, but by having the right context, the right map of reality, and then quickly adjusting its map or its actions to limit surprises.

If you’re already a student of my work at Organizational Physics, then you already have a very sound mental model to bring more “life” to your business:

  • Build Your Internal Map. Organisms cannot function—they cannot predict or minimize surprise—without a clear, actionable internal map of the business and its terrain. A sound business map must define reality and provide the necessary context for your people and AI agents to make the right decisions before the surprise hits.
  • Monitor Energy Gains and Drains. Responsive systems pay attention to energy gains (a symptom of integration or flow with the environment) and energy drains (a symptom of entropy or negative surprises, friction, or energy loss that must be recovered).
  • Manage to the Stage. A key element of minimizing surprises is to pursue the right activities for the right stage of business development. If you try to gather nuts in the winter rather than the summer, you are fighting reality and are doomed to failure.
  • Gather the Mass. In physics, mass is resistance to change. In business, you must harness this resistance. Without the continuous alignment of your Goals, Culture, Structure, Process, Data, People, and AI Agents to your core Strategy, you can’t build organizational momentum. When any of these elements become misaligned, they create friction instead of momentum, resulting in costly internal surprises.

The bullets above represent the fundamental physics of how your business actually works. To thrive in this era or any other, you must master and leverage each one. If these concepts speak to you, dive deep and learn more. That’s what a leader does.

Part VI: The Era of Efficient Inference

If the Free Energy Principle sounds too abstract to apply to your business, there is a very practical reason to pay attention: Efficient Inference.

We are entering the era of efficient inference. In AI, “inference” is the moment the model actually does the work to answer a question. The models that can generate the most accurate answers with the least amount of energy and compute will dominate those that rely on brute force.

Dr. Friston is the Chief Scientist at Verses AI, which is applying the Free Energy Principle to machine learning. The results are stark. In a recent benchmark test on logic and reasoning (the “Mastermind” game) against the massive DeepSeek R1 model, the Verses ‘Genius’ model was 245x faster and 779x cheaper to run per solution. It also solved 100% of the challenge games, whereas DeepSeek R1 solved only 45%.

I’m not affiliated with Verses, but I am calling out the trend: over the coming years, efficient inference will be the defining metric of success. The winning models, and the winning businesses, will be those that can consistently close the gap between prediction and outcomes with minimal energy.

This was always true for biology and for business, and it is now becoming true for AI.

In all three scenarios, the winner is the one designed to sense and respond like a living system—consistently and efficiently minimizing surprises against changing conditions at scale. That is the challenge, and the opportunity, in this era and the next.