How Block Scaled AI Agents to 12,000 Employees in Just 8 Weeks

Block, a fintech company, deployed AI agents to all 12,000 employees within eight weeks.

Reading guide: Block’s VP of Engineering recently shared how the fintech giant rolled out AI agents to every department in just eight weeks.

Small Start

The story began with a dissatisfied engineer, Bradley, who was frustrated with AI coding tools that only generated code snippets and wanted a tool that could automate complex development tasks.

OpenAI’s function calling feature was introduced, and Bradley’s team started automating parts of their development workflow. Success quickly turned into scalability challenges due to a lack of standards for integrating different APIs.

Anthropic introduced Block to a new technology called Model Context Protocol (MCP), an open standard for connecting AI agents with tools and data sources. Block became a founding partner of MCP and rewrote its internal AI agent, Goose, to be MCP‑compatible.

Unexpected Discoveries

While building Goose, the team realized the project could serve every employee, not just engineers. Non‑technical staff struggled with software installation and API key management, exposing the system’s developer‑centric design.

To address this, Goose was added to the company’s internal software control center for automatic installation and updates on all laptops.

Employees were allowed to choose from OpenAI, Anthropic, Google, and Meta models, giving each team flexibility and encouraging adoption.

Security was a major concern; Block, as a regulated fintech, needed strict safeguards and built an internal server team to manage tool access.

Making It Truly Usable

The team shifted from an engineer‑first mindset to a user‑first approach, simplifying installation by auto‑installing Goose via the internal software center.

They automated tool management: when a user asked about calendar events or Asana tasks, the system automatically enabled the relevant tools and disabled others.

They also added conversation summarization to keep long chats concise.

Unspoken Issues

Running many tools simultaneously slowed AI agents, prompting the development of automatic tool activation and deactivation based on user queries.

Building Support Systems

Block invested heavily in community support, creating two Slack channels—one for help and one for sharing projects—and hosting weekly webinars, Q&A sessions, and team‑specific training.

What People Actually Built

Security analysts now query data warehouses in plain English to detect fraud without writing SQL.

A sales operations employee re‑allocated 80,000 sales records in an hour instead of days.

An engineer used event‑management integration to identify and fix infrastructure failure patterns.

Non‑technical staff also built tools, such as an MCP server to streamline event sponsorship requests.

Big Picture

The Block experience shows that successful enterprise AI requires more than technology—it needs standards, easy installation, security, training, and community support.

Investing early in security, compliance, education, and user‑driven innovation is essential for scaling AI across an organization.