AI and Pointless

As I write this in March 2026, AI has played a very limited role in the design and development of the Pointless language, its documentation, and its website. In this document I'll describe the ways in which this may change, and the ways in which it will not.

I've been working on this project for around 8 years now; making engineering mistakes, applying my knowledge of algorithms to a variety of tricky problems, writing some pretty clever code, and going through many, many iterations of prototyping, testing, and revision. I love software development, and I consider this project to be a testament to my creativity and skill as a software engineer, built through 8 years of hard, human work.

In the process, I've built a language that I am very proud of, and that I think has real-world potential, particularly as an educational tool. As such, I'm starting to think about how to better document, test, and publicize what I've built. I think that generative AI could be helpful in this next phase of work.

How I have used AI so far

As of March 2026, essentially all of the code in this project has been designed and written by hand, with the exception being very small code snippets (things like "write a function to escape an HTML string in JS"). Until now, there have been fewer than 100 lines of AI-generated code in this project across the core language code, standard library implementation, documentation, tooling, and website. Any AI-generated code that is currently present was not added using agentic AI, but was instead manually reviewed and incorporated into the project.

I myself started using AI regularly about a year ago, slowly incorporating it into my work. Over the past six months, when working on this project, I've used generative AI to help me understand APIs and specifications of existing software, and get feedback on design decisions. Examples of questions I've asked AI include:

• What is the difference between AbortController and AbortSignal?
• What are valid characters in an ISO 8601 string?
• Why does {}[print("a")] = print("b") print "b" before "a" in Python?
• Are there any languages that include a break keyword for loops but not a continue keyword? (Lua)
• Which of these standard library functions (...) should be globals?
• I'm writing a standard library function that does (...), what should I name it?

I've also used AI to get feedback on some larger design choices, with questions like:

I'm thinking of allowing the arg keyword before a field name to be omitted and inserted implicitly by the parser, which would let me write code like this cities $ .population > 100000 as a shorthand for cities $ arg.population > 100000. Is this a good idea? (It was not)

How I plan to use AI going forward

• I won't use AI to generate any person-to-person communications (emails, comments, etc).

• I won't use AI-generated images anywhere in this project. I may use AI to generate SVG diagrams if it ever becomes good at that.

• I won't use AI to generate any articles or tutorials on the language website. I may use AI to help me scaffold my writing: for example, if I have a list of topics I want to cover I might ask AI for suggestions on the best order in which to cover them.

• I won't use AI to write any of the core language or standard library code (the code in pointless/lang and pointless/stdlib)). I know this code like the back of my hand and I want it to stay this way. If I ever decide to overhaul the language implementation (have it compile to WASM, say) then I might revisit this decision.

• I may use AI to generate code for language-related tooling (the language REPL has been a particular headache that I'd love to get some help with).

• I may use AI to help generate the HTML, CSS, and JS code for the langauge website.

• I will use AI to help write a (long overdue) test suite for the language.

• I will continue to use AI to get feedback when making language design decisions.

• I will use AI to help tighten up the written descriptions for standard library functions. I may also use AI (along with careful human review and editing) to help generate new documentation for standard library functions. I've found that AI is very good at writing clear and concise specifications.

• The question of language documentation is a little tricky, since it sits somewhere between a specification and a tutorial. I won't use AI to generate language documentation from scratch, but I may use it to revise draft content that I've written.

• I will use AI to lint my code and proofread my writing.

• I will continue to ask AI what to name standard library functions. It is the hardest problem in computer science, after all.