This is application development where a model does part of the work: reading a document, classifying an enquiry, drafting a reply for a person to approve. NUZM builds these with Make, n8n and BOLT where a no-code tool is genuinely sufficient, and in ordinary code where it is not.
What this includes
- A defined boundary: what the model decides, and what a person still signs off.
- Prompts kept in version control, so a change to one is reviewable like any other change.
- A record of what the model was asked and what it answered, which is the only way to debug it later.
- A cost ceiling, because usage-priced APIs fail expensively rather than loudly.
Is this the right choice for you?
Right for language-shaped work at volume: sorting, summarising, extracting fields from documents, drafting first versions.
Wrong where the answer must be exactly right every time. A model is a probability, not a calculation. Anything that has to be arithmetically correct should be arithmetic.
Worth knowing before you commit
No-code tools are excellent until the day you need something they do not do, and then the whole thing has to be rebuilt. We use them where the workflow is genuinely within their range and say so when a project is going to cross that line, rather than discovering it after you have paid for the first version.