How Forter Trained 200 Non-Technical Staff to Build AI Agents in Two Weeks

When Ben Maraney, a principal engineer at Forter, mentioned to peers at a tech conference that his organization had enabled 200 team members across research and development to build AI agents in just two weeks, the reaction was one of surprise—many considered it "impossibly far away." Yet Forter, an identity and fraud-detection firm, pulled it off by deliberately avoiding the hard parts of agent development.

Forter's core challenge was straightforward: the company needed to empower its analysts—many without coding backgrounds—to build intelligent systems that could research fraud patterns and customer behavior in real time. Rather than tackle the full complexity of modern AI systems, the team identified three key levers: purpose-built tools, multiple framework options, and the removal of organizational roadblocks.

Building a Shared Tool Library#

The foundation was Toolchain, an internal Model Context Protocol (MCP) server that exposed tools to agents through a unified interface. Analysts could browse available tools in one place, configure which ones an agent should access, and get connection details to hand off to their agent—all without getting confused by too many options or creating security gaps.

Creating new tools was designed to be accessible. Contributors cloned a repository, followed examples, and added a configuration file describing the tool's purpose, parameters, and usage patterns. The approach worked: the tool library grew from about 20 tools at the team's founding to nearly 100 by the end of the sprint.

One example of simplification: instead of building a retrieval-augmented generation (RAG) system from scratch, Forter exposed three tools wrapping Glean, an enterprise search platform already connecting to Confluence, Asana, Jira, Slack, and Salesforce. Agents could search for documents, fetch full texts, or request focused summaries—bypassing the complexity of chunking, indexing, and ranking that manual RAG requires.

Multiple Platforms for Different Needs#

The team recognized that agent use cases varied. Some analysts wanted interactive chatbots; others needed event-triggered workflows or scheduled runs. For chat, Forter offered two paths:

• LibreChat (no-code): An open-source chat interface that agents could connect to via MCP. Development was fast and tool usage was transparent in the UI, making it ideal for experimentation. But MCP connections could drop, and customizability was limited.
• Template repos (code-based): Developers could clone a repository with boilerplate UI and full source control, enabling deeper customization and subagents—though deployment was slower, especially for analysts unfamiliar with engineering workflows.