The problem is no longer building an agent, but the six you already have

The interesting thing here isn't the builder, it's the admission. A vendor whose whole pitch was "add our agents" is now shipping a product to fix what happens when you do. Agent sprawl has arrived faster than agent value, and the failure mode is not a bad model. It's two competent agents acting on the same customer from different pictures of them. If you run more than two agents anywhere, that's your near-term risk, and it doesn't announce itself in a metric; it shows up as a customer getting a chirpy upsell mid-complaint. The lesson generalises well beyond HubSpot: shared context is the thing to design for, not more agents. And treat the 350 hours as what it is, one customer, one narrow, well-chosen task with a clean data source. That's the shape of automation that actually works. The teams getting value are picking small, dull, well-defined jobs, not building a workforce.