Mintlify will document code with AI – but there is no substitute for the human factor

Mintlify will document code with AI – but there is no substitute for the human factor
Using Mintlify Writer to generate documentation. Select the code, then click Generate docs

Mintlify, a startup specializing in automated code documentation, secured $2.8 million in seed funding this week.

The company was founded in late 2021 by Han Wang and Hahnbee Lee, both of whom studied at Cornell University, and follows an earlier project called Figstack which claims to explain code, answer questions about it, calculate time complexity (how many times a statement executes), and translate code between programming languages.

Figstack, an earlier product from the Mintlify team

Mintlify is focused on documentation. Mintlify Writer is an extension for Visual Studio Code and JetBrains IntelliJ that will automatically document code using AI. “We never store your code but code does leave your machine,” says the documentation on GitHub.

Languages supported include Python, JavaScript, TypeScript, PHP, Java, C#, Ruby, Rust, Dart and Go.

Documentation is generated in the form of comments that conform to standard formats, so it can be further processed by tools like the open source Doxygen.

Mintlify Writer has over 34,000 installs and a five star rating on the VS Code Marketplace, a good start. “Like Github Copilot for docs … I would argue it saves me even more time to auto generate docs, than copilot saves me,” said one enthusiast.

The company has another product, just called Mintlify, that describes itself as a “continuous documentation platform.” It will link code to documentation located elsewhere and “identify when enough code is changed to make the documentation incorrect”.

In other words, it addresses the common problem of code getting out of sync with documentation.

A few brief experiments with Writer and VS Code showed impressive, though not perfect, results. Writer excels at generating comments in English that are mostly both natural and concise, though all too often it adds little value.

That said, in one case we tried where some XML is generated, Mintlify Writer delighted us by doing what looked like real research into the purpose of a method, possibly based on an internet search for the DTD (Document Type Definition) and what the XML is used for, but also disappointed by repeating the same sentence four times over, which also happened to be based on an incorrect assumption.

It is early days though, and no doubt the AI will improve. Mintlify Writer looks ideal for organizations that require developers to document code as they go, though it is also possible to clutter code with comments that state the obvious while not imparting information of any real use. The downside is that over-reliance on a tool like this (as with GitHub Copilot) could have a negative impact or even introduce incorrect documentation.

As ever, the human factor remains critical.