AI features
AI where it genuinely helps, rather than a chat box in the corner because the category expects one.
Most AI features get built because the technology was there, not because anyone needed them. The ones that stick start with a job somebody is doing by hand far too often.
So we find that job first. Then we build around what the model gets wrong: how someone corrects it, what happens when it fails, and how you keep the bill from climbing as usage grows.
And we measure it, so you know whether it's working instead of assuming.