AI Agents Now Code From Your Instructions
A new way of building software using AI agents is here, and it’s called spec-driven development. Instead of just telling an AI to write code and hoping for the best, this method focuses on writing clear instructions first. This approach aims to make AI coding more reliable and easier to manage.
JetBrains, known for its developer tools, has partnered with an instructor to create a course on this new workflow. This system is different from what’s often called “vibe coding.” Vibe coding is like giving a vague idea to an AI and seeing what it comes up with. Spec-driven development, however, is a more organized way to work.
How Spec-Driven Development Works
The core idea is simple: you start by writing a detailed plan in a markdown file or a long text prompt. This document explains exactly what you want the AI agent to build. The AI agent then uses this specific plan, or “spec,” to write the code. This method is said to be how many top developers already work.
The new course teaches you how to write these high-quality specs. It covers how to use them for both starting new projects and working with existing code. Paul Everett, a developer advocate at JetBrains, is leading the course. He highlighted that the traditional way of coding, where developers wrote code that turned into machine instructions, is changing.
Today, much of a developer’s time is spent writing the instructions that turn into code. The main challenge is translating complex ideas and choices about an application into a clear specification. AI agents can’t guess everything; they need precise context. This course aims to teach developers how to provide only the necessary information that the AI doesn’t already know.
Reducing AI Errors and Developer Strain
One common problem with AI agents is “hallucination,” where they make things up or produce incorrect information. Another issue is “context rot,” where the AI loses track of what it’s supposed to be doing over time. Spec-driven development helps reduce these risks. Specs allow you to guide large changes in the code with just a few sentences.
This method also helps keep the AI focused. It preserves important information across many AI sessions, improving “intent fidelity.” This means the AI is more likely to understand and match your original goals. Developers can also experience “cognitive debt” and fatigue when working with AI agents. This happens when an AI generates a lot of code, and it becomes hard for the developer to keep up.
Specs help reduce this mental load. They let you manage the project at the level of what you want to achieve, rather than getting lost in the details of the code itself. By working in small, controlled steps, developers can stay on track and avoid feeling overwhelmed. The course focuses on writing specs with the right level of detail and using them in a workflow that can be automated.
Building an Example Application
To show how spec-driven development works in practice, the course uses a project called “Agent Clinic.” This is a fictional application where AI agents can go for help with their own problems, such as hallucination and context rot. By building this complex web application, students will learn by doing.
The course instructor, Paul Everett, noted that many skilled engineers building complicated software already use detailed specifications. He sees this as an important best practice that offers significant value to developers. The goal is to make AI coding more predictable and efficient.
Why This Matters
Spec-driven development could significantly change how software is built. It offers a more structured and reliable way to use AI coding assistants. This approach helps ensure that the code produced by AI actually matches what the user intended. It also makes it easier for developers to manage complex projects and avoid burnout.
For businesses, this could mean faster development cycles and higher quality software. By clearly defining requirements upfront, teams can reduce the time spent on fixing errors and rework. The ability to control AI agents with precise instructions is a key step towards more effective AI-powered development.
Availability
The new course on spec-driven development is available now. It was developed in partnership with JetBrains. Details on enrollment and pricing can be found on the course platform.
Source: New course! Spec-Driven Development (YouTube)
