Building agents for revenue teams taught us the model matters less than the context

We are building Rimplo, an agent platform for revenue and GTM teams. A few design decisions turned out to matter more than we expected, and I’m curious how others here think about them. 1. We stopped picking a “best” model. Early on we debated which LLM to standardize on. Every few weeks a new model beat the last one on something, and different tasks favored different models. A model that writes good outreach isn’t always the best at reasoning over CRM data. So we went model-agnostic. Users can run their agents on models from multiple providers and switch whenever they want. Locking a team into one provider means locking them into that provider’s weaknesses too. 2. Context beat model quality, by a lot. A generic agent asked “which accounts should I focus on this week?” gives a generic answer. The same model with the CRM, product usage, call notes, and support history behind it gives an answer a VP of Sales would actually act on. That’s why we built a context layer first: a shared picture of the business that every agent works from. The model reasons, but the context is what makes the answer useful. 3. Vertical beats horizontal for trust. Revenue teams don’t want a blank canvas. They want an agent that already understands churn, expansion, and pipeline. So Rimplo suggests what to act on, and users can build their own agents in plain English for the rest. 4. Every send needs a human yes. Agents draft and prepare, but nothing reaches a customer or touches the CRM without approval. Trust compounds slowly and breaks fast. For those building vertical agents: are you seeing the same thing, where context matters more than which model you pick?