When you abstract AI tools too much, you risk losing the very specific problems your customers are trying to solve.
The core issue with AI products today is that LLMs aren't smart enough to fully understand every niche, real-world edge case on their own. Especially in messaging-based commerce, where workflows are messy, informal, and highly contextual.
The Inventory Problem
While building Sakura, I spoke with a customer who sourced inventory from a WhatsApp vendor. The customer never actually knew whether a product was available unless they asked the vendor directly.
In Sakura today, you can add inventory data to train the AI. But in this case, the customer didn't technically have inventory at all—availability lived inside another human's WhatsApp inbox.
There was no clean abstraction for this.
The FDE Approach
So instead of forcing this case into our core product and complicating the UX for everyone else, we took a Forward-Deployed Engineering (FDE) approach.
That meant:
- Embedding closely with this customer
- Building a dedicated solution layered on top of Sakura's core
- Learning directly from how their workflow actually operates
Why This Matters
We've seen this pattern repeatedly. Different customers confirm orders differently. Some follow strict flows. Others rely on loose conversation patterns.
You can try to build features that cover every nuance, but doing that often breaks simplicity for customers who don't need that complexity.