AI Teammates: how monday.com runs production AI agents on Amazon Bedrock

This article describes how monday.com runs production AI agents at scale on Amazon Bedrock, with agents functioning as full teammates in their engineering organization.

• Three-level AI engineering framework: assistants (pair programming), skills/sub-agents (reusable workflows), and multi-agent (fully autonomous end-to-end delivery)
• Agents treated as stable team members with profiles, managers, scopes, and performance scores across Slack, GitHub, and monday boards
• Architecture uses SNS, SQS, EKS, RDS, ElastiCache, EFS, and S3 to route events and manage agent state across three inboxes
• Five retrofits: evals before upgrades, markdown-based memory files, remote sandbox testing, automated PR Guardrails, and monday boards as shared state
• Morphex agent achieves 95% autonomous merge rate by combining deterministic guardrails, eval trajectories, historical revert rates, and sandbox outcomes
• Confidence-scored auto-merge removes humans only when system signals are strong enough to improve outcomes

monday.com demonstrates that production AI agents require tight integration with existing engineering infrastructure, rigorous evaluation systems, and human-centered governance rather than separate agent-specific layers.