What is an AI coding assistant?

Your phone keyboard guesses the rest of the word you're typing. Programmers have had something like that for their code for years. With artificial intelligence (AI), it has grown from finishing a word, to finishing a line, to making changes across a whole project. That kind of tool is called an AI coding assistant. It suggests, writes, explains, or fixes code, and a person decides what to keep.

The first assistants worked a lot like that keyboard. As a programmer typed, a suggestion appeared in pale gray text, and one key accepted it. Today's versions look at much more. They read the project around the cursor, so they can suggest several lines at once that fit what the code is meant to do.¹

Newer assistants also take requests in plain language. A programmer might type, "write a function that checks whether an email address looks valid," and get the code back in seconds. Sometimes less. That skips the slow part: looking up how a thing is usually done, and typing it out. It can add a short note explaining the code, too, which helps the next person who has to read it.¹

They help with the parts of programming that aren't typing, too.

• Explaining: they read unfamiliar code and describe what it does in plain words.
• Fixing: they suggest why something broke and how to repair it.
• Testing: they write small tests that check the code does what it should.

For someone joining a project that's new to them, the explaining alone can save a lot of reading.¹

The newest kind goes further still. You describe a change, and the assistant plans the steps, edits several files, runs the tests, and tries again when something fails. International Business Machines (IBM) calls this agentic coding: systems that can look through a whole project, change many files, run commands, and keep going until the task is done. It works like an AI agent, a tool that takes a series of steps toward a goal on its own.¹²

Under the hood, it's still a model predicting what code should come next. So code can look right and be wrong, or work today and fail on an unusual case. Programmers know this. In Stack Overflow's 2025 survey of developers, 84 percent said they use or plan to use AI tools in their work. But 46 percent said they don't trust the accuracy of what those tools produce. That's why the guidance for one widely used assistant tells programmers to understand every suggestion and review it carefully before they use it.³⁴

Careful review is what sets professional use apart from vibe coding, where people describe what they want and often don't read the code. Both use the same kinds of tools, often the very same ones, but only one keeps a person checking every change. The next time an app on your phone updates, you might wonder how much of it an assistant wrote. And if you ever try one yourself, who would check what it changed?