How We Think About AI Integrations

A pragmatic view on adding LLM features to real products — what we build, what we skip, and why.

Every client conversation this year eventually touches AI. Sometimes the request is specific ("we want a chatbot"), sometimes it's a hunch ("we should probably have some AI in this"). Both are legitimate — but they lead to very different projects.

Here is how we think about it.

Start with the workflow, not the model

Retrieval before fine-tuning, almost always

Structure the outputs

Evaluate like you would any other feature

What we skip

Have a project where AI might help? Get in touch.