5 more AI agent deployments, and what each one doesn't prove

Hi,

This week we added 5 new sourced use cases. The directory now has 45: https://agentsinthewild.co

This week's 5

1. Formula 1's agents cut data-source onboarding from 6-8 weeks to about 40 minutes of code generation

Agents turn a requirements document into pull requests for pipeline code and GDPR tags. Engineers approve each pull request. The post says agents handle 95% of onboarding tasks without human help.

The catch: The 40 minutes is code generation only. Deployment and review take hours more. The 95% figure is not defined.

→ https://agentsinthewild.co/use-cases/formula-1-agentic-data-onboarding

2. AWS Sales put 20+ agents behind one assistant, Field Advisor

Reps ask one interface, which routes the request to the right agent. It asks approval before CRM changes. AWS reports 120K+ prompts and up to 2 hours saved per rep per week.

The catch: "Up to 2 hours" is an upper bound, and AWS wrote this about its own tool.

→ https://agentsinthewild.co/use-cases/aws-field-advisor-sales-agents

3. Reactiv's three-agent scheduler cut merchant app setup time by 80%

A supervisor, an analytics agent and a builder agent draft mobile app updates from Shopify store data. Merchants approve before anything goes live.

The catch: The numbers are Reactiv's own, with no method given.

→ https://agentsinthewild.co/use-cases/reactiv-ai-scheduler-agentcore

4. AWS Professional Services: four migration agents, infrastructure code in minutes instead of 3-4 weeks

Agents handle intake, infrastructure code, governance and post-cutover monitoring across 300+ applications. Write actions need human approval.

The catch: The timing comes from AWS's internal tracking, and the post says results will vary.

→ https://agentsinthewild.co/use-cases/aws-proserve-migration-agents

5. HUB International rolled out Claude to 20,000+ staff and reports 2.5 hours saved a week each

Insurance broker HUB uses Claude Enterprise, Claude Code and custom API tools. It also reports 90%+ satisfaction in early use cases.

The catch: It is a joint press release. There is no method or baseline, and the timeline on the page is unclear.

→ https://agentsinthewild.co/use-cases/hub-international-claude-rollout

Pattern of the week

Draft first, human approves. Four of the five keep a person in the loop: Formula 1 pull requests, Field Advisor CRM edits, Reactiv merchant approval, and AWS migration write actions. The time saved is in the drafting, not in removing the reviewer.

Prompt you can copy

From the AWS Field Advisor entry: route a request to the right helper. This prompt is ours, not theirs.

You are a router for a sales team. Agents available: {agent_list_with_descriptions}
Rep request: {request}

Pick the one best agent. If two are needed, give the order.
If the request would change CRM data, say so and ask the rep to approve first.
If no agent fits, say so. Do not guess.

Know a real deployment?

If you've seen (or built) an agent in production with a public write-up, send it in. Vendor case studies are welcome; we label them. → https://agentsinthewild.co/submit

Thanks for reading, Agents in the Wild