People often talk about no-code and AI as if they were the same thing. They do different jobs, and the best systems use each for what it does well.
A simple way to picture it: automation is the hands, and AI is the judgement. The automation moves information reliably from one place to another. The AI reads something messy and decides what it means.
Where the rules stop working
Take a delivery company in Nairobi that receives requests on WhatsApp. Customers write in every style imaginable: "pls pick up from Westlands 2pm", a voice note, or a long story with the address in the middle.
A rules-based automation cannot read that reliably. A language model can. It turns the message into clean fields such as pickup point, time and urgency. From there, the automation takes over: it creates the record, assigns a rider and sends a confirmation.
Three jobs AI does well inside a workflow
- Extract. Pull details out of messages, receipts and documents.
- Classify. Decide whether this is a complaint, an order or a question, and route it.
- Draft. Write a first version of a reply for a person to approve, or send it directly when the risk is low.
Where people go wrong
The common mistake is giving AI work that a simple rule could do. If you need to add 7.5% VAT, use a calculation. A model might get it right most of the time, but "most of the time" is not good enough for money. It also costs more and runs slower.
Use this test: if you can write the rule down, write the rule. If a person would have to read and think, consider AI.
Keep a person where it matters
AI will sometimes be wrong while sounding completely sure. So check its output before it reaches anything important. Confirm that required fields are present, send low-confidence cases to a human, and keep a log so you can see what happened. For refunds, debt conversations or anything legal, a person should review the result.
Start small
Take an automation you already have and add one AI step where someone currently has to read and sort. Run it beside your current process for a week, compare the results, and only then let it run alone.