How ABC Legal turned every employee into a builder with Claude Managed Agents

Summary

ABC Legal, a U.S. legal document delivery firm, rolled out Claude Enterprise to its 1,100 employees and observed rapid adoption of self-built automations across operations, finance, marketing, and compliance teams. CTO Brandon Fuller then introduced Claude Managed Agents to move those automations off individual desktops into a shared, always-on cloud environment with centralized observability, billing, and audit surfaces.

The decisive step was to treat every agent as structured configuration stored in Git repositories. A starter kit supplied two templates—one for event-driven agents triggered by incoming jobs or court returns, the other for scheduled runs—each containing a JSON config, Markdown system prompt, credential references, and deployment scripts. Non-developers cloned the repository, described the intended task to Claude Code, and received a complete, reviewable agent definition. Changes merged through pull requests, giving every agent version history, rollback capability, and an explicit audit trail without requiring the builders to write code.

Within a month the company had more than fifty production agents. Examples include an EvidenceChain Delivery Agent that extracts court records and posts PDFs to customer FTP servers, an eFiling Rejection Diagnoser that diagnoses court rejections and posts findings to Slack within a minute, and a finance agent that parses remittance emails and prepares NetSuite entries for one-click approval. Most agents post recommendations or completed work to Slack threads, where emoji reactions and replies supply labeled feedback.

A three-role feedback loop converts those reactions into versioned updates: a Harvester labels the data, a Tuner proposes configuration changes via pull request, and a fourth agent applies only human-approved changes to production. Cost tracking shows spend rising during rollout then declining as agents move from larger models to cheaper, faster ones once evals confirm performance parity with human operators. Each agent reports its own efficiency ratio—value delivered versus compute cost—allowing the team to shift proven agents from supervised recommendation mode to unattended automation while retaining the same measurement framework.

The approach keeps infrastructure responsibility with the managed platform and leaves prompt logic, tool selection, and outcome evaluation under ABC Legal’s control, enabling continued expansion of agents across document workflows without expanding the central development team.

Why it matters

This article provides a highly actionable blueprint for product teams and builders to deploy scalable, observable AI agents using a GitOps approach. It demonstrates how to empower non-technical staff to build automations while maintaining centralized governance, which is highly applicable to Dutch enterprises scaling AI.