Dieses Fallbeispiel ist derzeit nur auf Englisch verfügbar.
Rocket Money
Rowan personal finance agent
Rocket Money builds Rowan, a text-message finance agent on Claude, with commits up from 11 to 128 a month
Ausgangslage
Rocket Money wanted an agent that cancels subscriptions and negotiates bills for users, with an accuracy bar of 100%.
Vorgehen
Requests pass through a Haiku 4.5 classifier to single-purpose subagents. Routine tasks such as cancellations became fixed code, with an agent that repairs the code when a website changes. The team of about 23 people, 10 of them engineers, wrote code with Claude Code and plans a phased private rollout in the second half of 2026.
Kennzahlen
Erkenntnis
Anthropic reported near-zero hallucinations during the Rowan beta. A phased private rollout was planned for the second half of 2026.
Kategorien
- Größe
- Großunternehmen
- Branche
- Finanzen und Versicherungen
- Unternehmensfunktion
- KundenserviceIT und F&E
- Innovationsart
- Produkt- und Dienstleistungsinnovation
- KI-Technologie
- KI-AgentenGenerative KI
- Einführung
- Anbieterlösung
- Nachweis
- Fallstudie des Anbieters
Quellen
Quellenzusammenfassungen
Anthropic describes how Rocket Money built the Rowan agent on Claude. It reports near-zero hallucinations in beta and commits rising from 11 to 128 a month. The Rowan team had about 23 people.
Ausführliche Angaben finden Sie in den Originalquellen.