English as Code and the End of Drag-and-Drop Thinking

For as long as there has been code, the tech industry has been obsessed with making software easier to build. Low-code, no-code, drag-and-drop interfaces, all of it was designed around one assumption: digital systems will only understand code or structured LEGO blocks on top of it. But that assumption is now breaking down.

I have been saying for years that drag-and-drop is a weak crutch. It feels convenient, but it is not powerful enough. As AI begins to understand the lingua franca of businesses, we must stop asking how to make software easier for humans to assemble, but rather how we control intelligence at scale by expressing what we want in natural language.

See also: Vibe Coding: The New Literacy for the AI-Native Software Generation

Intelligence has always been controlled by written code

Throughout history, when humans have needed to control intelligence at scale, we have written down the rules. If I need to instruct one person, I can simply tell them what to do. If I need to instruct a thousand people, I cannot have a thousand separate conversations. I write it down, put it in a handbook or a guide. That written instruction becomes the governing logic.

At the largest scale, this is religion. Billions of people are guided by written texts. Those texts define moral code, ethical boundaries, and acceptable behavior. Individuals may interpret them differently, but the written word becomes the control mechanism. At the level of a nation, it is the constitution. Hundreds of millions of people operate within a written framework that defines rights, responsibilities, and limits. If outcomes need to change, the written rules change.

At the level of business, the same pattern holds. A company with ten employees may operate informally. A company with ten thousand employees must write its processes down. Policies, procedures, and operating manuals — these are not bureaucracy for its own sake; they are mechanisms to create consistency across distributed intelligence.

When McDonald’s wants to improve the taste of its fries, it does not retrain every cook individually. It changes what is written down. It adjusts the documented process, the salt level, the temperature, and the sourcing standard and distributes that instruction globally. Even if a world-class chef is working in one location, that person must follow the written code.

This is how humans have always controlled intelligence: by codifying behavior in language.

And in the process, humans have retained control of their collective future.

See also: Vibe Coding: How AI-First Workflows Are Redefining Customer Data Engineering

From prompt engineering to English as Code

Early AI relied heavily on prompt engineering. We treated prompts as clever tricks to guide model behavior. But prompts are fragile, model-specific, change when the model changes, and just aren’t durable governance. This is similar to how you explain a task to someone, depending on their experience level and background.

Low-code and no-code platforms are also transitional. They abstract complexity for humans, but they do not solve the problem of controlling autonomous intelligence.

What will survive is something much simpler and much more powerful: English as code. When I say code, I do not mean cryptic symbols that only trained developers can understand. I mean code in the original sense. Think, code of conduct, moral code, ethical code. Written instruction that defines behavior.

AI will ultimately be governed the same way humans are governed in organizations. We will write down what we want. We will define objectives, boundaries, and acceptable actions in natural language. And AI systems, regardless of the underlying model, will be required to follow that written instruction.

The model may change every two months, the intelligence layer may evolve rapidly, but the written code remains.

See also: Vibing on AI Governance

The happy path and the tribal knowledge

There is another dimension to this that most people underestimate. In any organization, there is a documented process, and there is tribal knowledge.

The documented process defines what to do when the world behaves as expected. The problem is that the world rarely behaves as expected. Someone slips on the floor. A customer escalates unexpectedly. An edge case appears that no one thought to encode. That is the moment when implicit judgment takes over.

Humans rely on common sense and cultural alignment to navigate those moments. You do not need a written instruction that says, “Help a customer who has fallen.” You help because it aligns with shared values. AI does not have that implicit alignment, and even in cases where it does, it does not know how far it needs to go without approval.

If we want AI to behave reliably inside enterprises, we must progressively codify not only the happy path but also the judgment calls that today live in human heads. When a system encounters an exception and a human steps in to resolve it, that decision should not disappear into memory. It should be captured as English code so that the next time the scenario occurs, the AI knows how to respond.

It is not enough to train a model and hope it behaves correctly. That is equivalent to hiring employees without writing down the company’s operating principles and assuming culture will magically align. At small scale, that might work. At enterprise scale, it does not.

Why drag-and-drop thinking fails

Drag-and-drop tools assume that the primary challenge is making system construction easier for humans. But as AI becomes the primary executor of workflows, the constraint shifts. Why? Because we are now optimizing for predictable, auditable behavior from autonomous systems.

A visual workflow builder might make a simple process easy to assemble, but it fails at providing affordance for nuances that invariably appear in real-world scenarios. Those nuances often are not easily encoded as workflow charts but can be written down clearly in natural language.

Natural language is the only medium flexible enough to encode both structured process and nuanced policy. It can describe logic, but it can also describe values. It can define what to do and what not to do. It can articulate not just technical constraints, but moral and operational ones.

In that sense, English as code is not just a programming concept. It is a moral and ethical control layer for AI.

The durable layer in a rapidly changing stack

AI models are improving faster than human hiring cycles. What is state-of-the-art today may be obsolete in six months. If your business logic is embedded in a specific model’s behavior, you are building on sand. The only stable layer in this stack is the written intent.

When we treat English as executable business logic, we decouple governance from model capability. We can swap models as they improve, just as companies replace employees over time, without rewriting the foundational principles of operation.

The future of AI in the enterprise is recognizing that intelligence, whether human or artificial, must be governed by clear, written code.