Skip to content
Agents tracked: 502 Releases (7d): 613 Agent status: 4 with issuesSince last limit reset: Claude 7d · Codex 5d For AI agents: Markdown Updated Nov 18, 2026
Case study LeagueResults as reported by Anthropic

League halved product cycle times with Claude Code and overnight agent teams

League, which builds member health apps for health plans and providers, rolled Claude out company-wide and moved its engineers to Claude Code. Small pods now ship production code behind feature flags in one sitting, larger builds run overnight through an in-house multi-agent tool, and a Claude skill handles vendor security assessments. Anthropic reports cycle times from idea to pull request cut in half.

The problem

League works with protected health information, so every new tool has to clear a high compliance bar. AI assistants gave staff only "tiny little pockets of efficiency", security reviews backed up for about three weeks, and customer implementations ran on traditional healthcare timelines with two-week sprints.

How they did it

  1. Do the security groundwork before scale. League kept patient data in a separate environment Claude cannot reach and set strict limits on what agents could do while controls were put in place.
  2. Force the new pace with a short event. League's Accelatron asked teams to take real backlog items from backlog to production in 48 hours, not to build hackathon demos.
  3. Replace sprints with micro-sprints. Pods of three or four people scope, build and land production-ready code behind a feature flag in a single sitting with Claude Code.
  4. Send larger builds to a lead agent that splits the work and runs a team of agents in parallel overnight; an engineer reviews the result the next morning. League built its own tool (Swarm) for this; Claude Code's experimental agent teams give a similar lead-plus-teammates setup.
  5. Keep review from becoming the bottleneck with a Claude-based bot that triages every pull request by size of change and risk, for example a Claude Code GitHub Action that labels each PR before a human picks it up.
  6. Package repeat work as skills. League's vendor security risk assessment runs as a Claude skill, and a human signs off every result.

From the official docs

{"env": {"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"}}

Results as reported by Anthropic

  • Cycle times from idea through pull request cut in half in the four months after the Accelatron, as reported by Anthropic.
  • AI-authored code up from roughly 70% to 98%, tracked on League's engineering metrics platform, as reported by Anthropic.
  • Engineers merge roughly two to three times as many pull requests per week as before the rollout, as reported by Anthropic.
  • Vendor security risk assessments down from multiple weeks to 15 minutes, with 49 of the last 53 validated as safe by Claude and then signed off by a human, as reported by Anthropic.
  • Security team response times down from 3+ weeks to a few hours, as reported by Anthropic.
  • 98% adoption of Claude, up from ~80% when the company-wide rollout began in March 2026, as reported by Anthropic.
  • One customer implementation delivered two months ahead of schedule, and more than 60 finance processes automated, largely by the finance team itself, as reported by Anthropic.

These figures are as reported by Anthropic in Anthropic customer story ↗. Anthropic sells the product used here. AgentGid did not measure or independently verify them.

Takeaway. The speed came from changing the work unit (one-sitting micro-sprints, overnight agent runs, triaged review), not from handing engineers a new tool and keeping two-week sprints.
AgentGid's take

This fits a product company with feature flags, solid CI and engineers willing to review agent output every morning; the overnight multi-agent part is League's own tool on top of Claude, and Claude Code's agent teams are still experimental and use far more tokens than one session. Regulated teams should copy the order of work: data isolation and agent limits first, rollout second. Per AgentGid's Claude Code page (pricing checked Nov 4, 2026), Claude Code needs a paid plan: Team standard seats are $20 per seat a month on annual billing ($25 monthly), premium seats $100 ($125 monthly), and Enterprise seats $20 per seat a month annual with a 20-seat minimum plus usage at API rates. Pro starts at $17 a month annual. Every figure is from Anthropic's own customer story, with no publication date and no independent measurement. For an open-source route to parallel coding agents, see OpenHands or OpenCode in our catalog.

The agent used here

Similar use cases

Guides