Stewart Rutledge2025-04-25

Using AI To Help Write Specifications

When we’re developing M3 integrations, H5 apps, and scripts, we’re always looking for ways to improve our workflow, speed up delivery, and continually raise the quality of what we ship.

Like everyone else, we’ve been exploring how AI can help with that. We’ve shared before how we’ve used AI to generate H5 scripts, and we’ve even started experimenting with tooling to build out boilerplate H5 apps faster—with help from AI.

Recently, though, we’ve been digging into how AI tools like ChatGPT can help improve the quality of specifications. Writing specs is hard—finding the right balance between being succinct, descriptive, technical, and easy to understand isn’t easy. So far, we’re pretty impressed with what we’re getting.

Here’s a (contrived) example:

“Customer order interface. We have an external webshop that needs to get orders, items, and customer data from M3 when events occur. For this, we need events on MITMAS, OCUSMA, and OOHEAD/OOLINE. When an event for the corresponding table happens, we’ll run a MEC mapping that creates Webshop XML and passes it to ION, which uses the webshop’s API to create the record.”

From that, we get a pretty solid spec output (screenshot attached).

As always, the more detail you give, the better the result. But we’ve found that even working from existing specs—especially when the original author isn’t available—AI can really help restructure and clarify things.

Anyone else using AI to help write specifications for M3?