Citi Launches Arc Platform to Scale AI Agents Enterprise-Wide

Citigroup has officially launched Arc, a centralized platform designed to build and scale AI agents across the bank's operations. The announcement marks a significant shift from experimental AI pilots to industrialized infrastructure for autonomous agent deployment.

According to the bank's official announcement, Arc will initially be used by Citi developers to construct agents for specific internal use cases. Access will expand over time, with employees seeing agents integrated into everyday workflows to surface deeper insights from data and reduce repetitive tasks.

"For the first time, we can deploy embedded AI agents at enterprise scale across every business line, every geography, every function," David Griffiths, Citi's CTO, said in a video announcing the platform. "The real game changer is that we've built industrialized infrastructure to make AI agents a core part of how we serve our clients."

The platform functions as a centralized operating system for AI agents, according to Axios reporting. This architecture allows employees and managers to monitor agent behavior and stop tasks if needed, which helps prevent agents from going rogue (a legitimate concern when autonomous systems handle financial data).

Citi's move follows a broader modernization push. In a Q1 2026 earnings call last month, CEO Jane Fraser highlighted the company's efforts to deploy AI to increase revenue, improve client experiences, and drive process improvements. The bank appointed former Google executive and financial services IT veteran Brian Saluzzo as CIO, effective in March, to help the bank further scale AI capabilities.

The foundation for Arc already exists. More than 80% of the bank's 180,000 employees with access to Citi AI tools regularly use the technology. Most employees have also completed prompt training to better understand how to gain the most value from the technology, according to the announcement.

Agents built on Arc can compile portfolio data, analyze broad market trends, and test scenarios. The platform links agents and use cases into one central location, creating a unified view of AI activity across the organization.

Wall Street is racing to deploy agentic AI securely. Morgan Stanley's wealth management business is developing several AI agents to automate routine tasks, freeing up employees to spend more time with customers. BNY employs AI-powered digital employees, including digital engineers that fix low complexity code.

Nearly 6 in 10 banking executives believe AI agents will be fully embedded in risk, compliance, and audit functions, as well as fraud detection and transaction monitoring, within the next three years, according to Accenture's Top Banking Trends for 2026. Another 56% expect the technology to be broadly adopted in "know your customer" functionalities, credit assessment, and loan processing.

The physical reality of this shift matters. Employees will no longer just query a chatbot and wait for a response. Instead, they'll trigger agents that execute multi-step workflows—pulling data from multiple sources, organizing it, and drafting accompanying communications. The difference between a chatbot and an agent is like the difference between asking a librarian for a book versus having them find it, check it out, and mail it to your desk.

Security remains paramount. Banks are racing to offer the best AI models to employees without compromising on safety. Arc's centralized architecture allows Citi to maintain oversight while scaling agent deployment. That means employees and managers can monitor agent behavior and stop tasks if needed.

Other companies are building similar infrastructure. Snowflake says its platform—called Project SnowWork—can autonomously build pitch decks by pulling data from multiple sources, organizing it, and drafting an accompanying email. Sycamore is an agentic AI operating system designed to build, deploy, and orchestrate agents. Founded by former Atlassian CTO Sri Viswanath, the company raised $65 million in seed funding in March.

Whether Citi's Arc becomes the industry standard or just another proprietary tool depends on execution. The technology exists. The question is whether employees will actually trust agents with critical workflows, and whether the bank can maintain the infrastructure without creating new points of failure.

For now, the rollout begins with developers. The broader bank will follow over time. Whether that timeline accelerates or stalls depends on how smoothly the initial deployments perform—and whether the agents actually deliver value beyond novelty.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn