Software development
Coding with AI: from understanding code to a reviewable change
A practical workflow for AI-assisted coding: provide context, keep changes focused, review the proposal and run the checks that matter.
Clarify the problem before generating code
AI-assisted programming starts with understanding the problem. For a bug, describe the current behavior, expected behavior and reproduction steps. For a new feature, define inputs, outputs and constraints. A code snippet without context can lead to a suggestion that does not fit the project’s architecture.
Begin by asking the model to explain the relevant code and identify uncertainties. This helps you see whether it understands the relationship between components before proposing changes. Remove secrets and user data from the material you share; a focused example is usually easier to reason about than a large, unrelated collection of files.
Ask for one small, specific change
Combining an architecture rewrite, a bug fix and a visual redesign in one request makes the outcome harder to review. Choose a clear goal and define the scope. If the public API or existing behavior must stay compatible, say so explicitly.
Ask for the reasoning behind the proposal. The model should explain which cause the change addresses and what assumptions it relies on. A clear explanation does not replace running the code, but it makes incorrect assumptions easier to spot before they become part of the implementation.
Try this example
This function fails on empty input. It should return an empty array. Propose the smallest necessary change while keeping the function signature and behavior for valid inputs unchanged. Then write one test for empty input and one for a typical input.
Review AI-generated code like any other change
Read the proposal and confirm that its libraries, functions and configuration options exist in your project’s installed versions. A model can suggest an outdated API or a function that does not exist. The documentation for the installed version and the patterns already used in the repository are useful starting points.
After applying the change, run relevant tests, type checks and a build when appropriate. If the change affects an interface, inspect the behavior in a browser too. A passing test confirms the behavior it covers; it does not establish that the entire system is correct.
Make the result understandable to the next developer
A useful change description answers three questions: what was wrong, what behavior changed and how was it checked? Record those answers in the change summary or pull request. If a limitation remains, describe it briefly and concretely.
Use Ditfa conversations to explain code, compare approaches and review a snippet. Direct file access and tool execution depend on the environment and available features. In a text conversation, provide the relevant code and test output yourself so the discussion stays grounded in evidence.
Put it into practice with a real task
Ask a specific question, provide the context and improve the answer one step at a time.
Start a conversation in DitfaKeep exploring
What is Ditfa? A workspace for thinking, writing and building with AI
Meet Ditfa: a workspace for AI conversations, writing and code exploration. Learn how to turn a clear question into a practical workflow.
Read article
How to write better AI prompts: a practical guide with examples
Get more useful AI responses with a clear goal, enough context, explicit constraints and an output format. Includes a sample prompt and review checklist.
Read article
Choosing an AI model: compare quality, speed and cost
Choose an AI model for your actual task. Compare response quality, speed, cost and available features using a small set of real work samples.
Read article