Halter Turns AI on Its Own Engineers as Virtual Fencing Platform Automates Incident Response

The New Zealand agritech company behind GPS collar-based virtual fencing has built an autonomous AI agent called Clank to investigate system faults, diagnose root causes and draft code fixes before engineers log on — cutting 215 hours of manual work across a platform carrying more than 16,000 monitoring alerts, on which live animals depend.

Halter’s farmer-facing proposition is already well established: GPS-enabled collars and a mobile app allow cattle to be managed without physical fences, with live data on location, wellbeing and grazing behaviour informing pasture decisions. One customer has relocated his family to France while continuing to run his farm remotely.

The newer development sits in the back office. Clank runs on Amazon Bedrock using Anthropic’s Claude model, integrating with Slack, GitHub and Linear. When a CloudWatch alert fires, a webhook launches an ephemeral task on AWS Fargate via Amazon ECS, which pulls the relevant logs and system data before passing them to Claude. The model correlates events, identifies root causes and generates a structured response. In many cases it proposes code changes and opens pull requests for human review.

The reliability argument is the one that matters commercially. Craig McAlpine of Halter put it directly: there is an animal on the other end of the technology, making robustness and scalability non-negotiable, and the company selected Bedrock for the reliability and data privacy it required before deploying Clank.

For a sector where platform uptime translates directly into animal welfare and grazing outcomes, the question now is whether that trust holds as Halter extends its connected farming model beyond New Zealand.