Resources
Integration guides
How a model gets into software that already exists, written for the person who has to approve it rather than the one who writes the code.
Latest
6 pieces
9 min readWhat it actually takes to add AI to software you already runThe work is mostly endpoints, data handling and ownership. Very little of it is the model.
7 min readAPI integration or a private model, and how to tell which you needWhere the data is allowed to sit decides this, long before any benchmark does.
6 min readThe questions to settle about your data before anybody buildsRetention, residency and who reads the logs. Retrofitting these means doing the integration twice.
8 min readConnecting a legacy CRM to a current model without touching its codeAn encrypted bridge and a middleware layer, for systems whose original developer is long gone.
What an AI integration actually turns on
A model going into software that already exists is mostly not a modelling problem at all. The endpoint you call and the place your data is allowed to sit decide more than the model choice ever does.
- Which endpoint
- A hosted API is faster to ship and an open weight model on your own hardware keeps the data in the building. Where the data is allowed to sit decides it, and benchmark scores rarely come into it.
- What happens to the data
- Retention switched off on a hosted endpoint is what keeps your inputs out of training sets. Residency and who can read the logs get settled before anybody writes code, because retrofitting them means building twice.
- Who holds it afterwards
- The keys and the middleware should end up with you rather than with the agency. An integration nobody but the agency can change is a dependency wearing a different name.
Next stepLet AI do the repetitive
Let AI do the repetitive
half of the job.
Data entry, answering the same tickets, chasing numbers between systems. We automate the parts that repeat. Your team keeps the parts that need judgement.
