Case Studies

Citigroup deployed agentic AI capabilities in its proprietary Citi Stylus Workspaces platform, piloting with 5,000 employees across the US, UK, India, and Canada in September 2025. The platform, powered by Google Gemini and Anthropic Claude models, enables employees to complete complex, multi-step tasks—such as client research, profile building, and translation—with a single prompt, eliminating the need for separate human direction at each stage. The initiative represents a shift from AI as a support tool to autonomous agents capable of executing workflows independently.

Block deployed codename goose, an open-source AI agent powered by Claude 3.5 Sonnet on Databricks, saving 75% of engineers 8-10+ hours weekly and enabling 4,000 of 10,000 employees to independently solve data problems. The deployment contributed to a strategic workforce reduction from 10,000 to 6,000 employees while doubling gross profit per employee from $500k to $1M, with projections to reach $2M by 2026.

Citigroup deployed agentic AI capabilities in its proprietary Citi Stylus Workspaces platform, piloting with 5,000 employees across the US, UK, India, and Canada starting September 2025. The platform, powered by Google Gemini and Anthropic Claude models, enables employees to complete complex, multi-step tasks—such as client research, profile building, and translation—with a single prompt, eliminating the need for separate human direction at each stage. The broader Stylus platform reached approximately 180,000 employees across 84+ countries, with AI tools used 7 million times in 2025.

TP ICAP, a global financial markets infrastructure provider, deployed a generative AI compliance solution on AWS in just four weeks, reducing regulatory document review time from approximately one month to minutes. The system, built using Amazon Bedrock and Amazon Textract, enables non-technical compliance staff to upload documents and receive instant assessments, transforming a manual process that previously consumed weeks of analyst time.

Robinhood deployed FinCrimes Agent, a generative AI solution built on Amazon Bedrock, to transform financial crimes investigations. The multilayered system uses Anthropic's Claude Haiku and Sonnet models alongside DeepSeek to synthesize customer and transactional data into concise investigative summaries. The deployment achieved a 20% cumulative efficiency gain in investigative workflows, allowing analysts to identify illicit activity with greater focus and accuracy while maintaining rigorous standards for explainability and regulatory compliance.