Intent Over Syntax

For the first six decades of software development, learning to program meant learning a formal language — a precise, unforgiving notation where a misplaced semicolon could break an entire system. Syntax was the gate. If you could not speak the machine's language fluently, you could not build. Vibe coding reorganizes this relationship: the gate is no longer syntax, it is intent. What you want to build must be specified precisely enough for an AI to act on it. That precision, it turns out, is its own demanding discipline.

What Syntax Is and Why It Mattered

Syntax is the formal grammar of a programming language — the exact rules governing how statements must be written for a compiler or interpreter to understand them. Python requires consistent indentation. JavaScript distinguishes between == (loose equality) and === (strict equality) in ways that trip up beginners for years. SQL has a mandatory clause ordering: SELECT before FROM before WHERE. These rules are arbitrary human conventions that exist because machines need unambiguous instructions. Before AI code generation, the only path from an idea to running software ran through syntax. A product manager who wanted a report could describe it in plain English all day; without the syntax knowledge to implement it — or the budget to hire someone who had it — nothing ran. Syntax was not just a technical detail; it was a gatekeeping mechanism that determined who could build. Mastering syntax took real time. A working knowledge of Python takes most learners 50-100 hours of deliberate practice before they can write small programs reliably. Full professional fluency — handling edge cases, understanding idiomatic patterns, knowing the standard library — took years. The investment was real, and it was the barrier to entry.

What does precise intent specification look like in practice? Compare these two prompts for the same task: Vague: 'Make a thing that handles user logins.' Precise: 'Build a login form that accepts an email address and a password. Validate that the email is properly formatted before submitting. Send a POST request to /api/auth/login with the credentials as JSON. If the response status is 200, store the returned JWT token in localStorage under the key auth_token and redirect to /dashboard. If the response status is 401, display the message Invalid credentials. Try again. under the form. For any other error status, display A server error occurred. Please try again later.' The first prompt will produce something — probably a generic placeholder. The second prompt specifies every decision point: inputs, validation rules, the API contract, the success path, the two distinct error cases, the storage mechanism, and the user-facing messages. An AI receiving the second prompt has almost everything it needs to produce a correct first draft. The skill of writing the second prompt is not a matter of knowing Python or JavaScript. It is a matter of thinking through a problem completely — anticipating edge cases, understanding data flows, making decisions about behavior before writing a single line of code. This is, in fact, what experienced software architects do before they write code anyway. Vibe coding surfaces that intellectual work as the primary task.

The New Skill: Rigorous Ambiguity Elimination

Natural language is inherently ambiguous in ways programming languages are not. 'Sort the list' does not specify: ascending or descending? Alphabetically or numerically? What happens with ties? What if the list is empty? What if it contains mixed types? A programming language forces you to resolve every ambiguity — you cannot write Python that half-sorts a list. Natural language lets you slide past ambiguities without noticing. This is vibe coding's central intellectual demand: you must impose on your natural-language descriptions the same completeness that a programming language enforces by syntax. Practitioners develop a mental checklist for prompts: What are all the inputs? What are all the valid outputs? What are the error conditions and how should each be handled? What should the system do at its boundaries? What assumptions am I making that I have not stated? Running through this checklist before prompting — and iterating when the AI's response reveals an ambiguity you missed — is the core practice of skilled intent specification.