How a 1,100-Person Legal Company Scaled to 50+ AI Agents in a Month

ABC Legal, a 1,100-employee US legal document delivery company (plus 50-person sister company Docketly), went from a Claude Enterprise rollout to 50+ production agents in a single month. CTO Brandon Fuller built a 15-person steering committee that shipped working agents within the first week, then scaled across service of process, eFiling, appearance counsel operations, marketing, compliance, and finance. By July 2026, roughly 310 employees used Claude daily, and agents covering specific human tasks had driven cost reductions of about 50% — before any heavy optimization. Named examples include an AI code reviewer checking pull requests across four codebases, an automated EvidenceChain delivery-reporting agent, an eFiling rejection diagnoser, and a compliance review agent nicknamed “Charvis” that agrees with the human compliance team’s judgment 98% of the time. The company defines agents as code stored in git repositories, uses a “harvester-tuner” pattern to refine them via Slack feedback, and builds human-in-the-loop trust before granting autonomy.

What stands out is where the agents ended up. Anthropic’s own usage analysis, published a few weeks earlier in how people are using Claude Cowork, found that across 1.2 million Cowork sessions spanning 600,000+ organizations, software development accounted for only 8.7% of usage — business process and operations was the largest category at 33.4%, with content creation second at 16.4%. Anthropic framed that spread as evidence of “the work around the work,” tasks that touch nearly every role but rarely show up in a job description. ABC Legal’s own agent footprint — marketing, compliance, finance, and delivery-reporting alongside a single code-review agent — is close to a live confirmation of that pattern inside one mid-size professional-services firm.

For consulting audiences advising clients on agentic AI rollouts, ABC Legal is a rare numbers-backed case: a steering committee, a defined build pattern, and a measured cost outcome, not a pilot with a press release attached.