Deep Dive
One prompt, a feature shipped
The Ramp AI lead describes Astra as a coworker you can hand off vague tasks to and check back in 10 or 15 minutes later. He submitted a single feature request: implement routing strategies controls by API key granularity. No step-by-step instructions. Astra spent 12 minutes building it, then another 15 minutes validating the changes. The model found the exact settings menu, made the changes, and even tested on mobile view without being asked. The speed and independence impressed him most — Astra asked for nothing beyond that initial goal statement.
The roadmap: customer-facing automation
Ramp's bigger vision is getting computer use into customers' hands. Today they fetch receipts from merchant websites by manually prompting the model on where to click. Ramp wants to test Astra against a longer tail of merchant sites to see if the model's general web understanding is strong enough to eliminate those manual prompts. If it works, customers could trigger receipt pulls without Ramp having to hardcode every merchant's UI path. This moves computer use from an engineering convenience to a customer feature.