monday.com's Sphera AI agents boost PR throughput 50% on Bedrock
90% of builders use AI coding tools monthly, per-engineer PR throughput up 50% at monday.com.
monday.com details its production AI agent system, Sphera, running on Amazon Bedrock. The journey spans three engineering levels: L1 (assistant pair programming with Cursor/Claude Code), L2 (reusable sub-agents driving a 50% PR throughput jump), and L3 (fully autonomous multi-agent delivery). At the center is Atlas, an agent with a profile, manager, and performance score—treated as a real teammate on the same Slack, monday.com, and GitHub backlog. The system handles three inboxes (Slack @mentions, monday item assignments, GitHub PR reviews) through a unified event pipeline.
The architecture relies on AWS services: SNS fans out events to per-team SQS queues, consumed by the monday Builders CoWORK on EKS. Each event resolves to an agent runner pod using the Claude Agent SDK wrapped in monday-agent-sdk for provider neutrality and cold-start mitigation. Amazon Bedrock handles model calls, while Secrets Manager, RDS, ElastiCache, EFS, and S3 support session state and storage. The confidence-scored merge play and durable replay (re-running last day's events against patched builds) enable safe autonomy in a decade-old codebase with millions of paying users.
- 90% of monday.com builders use AI coding tools monthly, up from ~50% a year ago.
- Per-engineer PR throughput increased by more than 50% using L2 reusable sub-agents.
- Architecture uses SNS/SQS fan-out, EKS, Bedrock, and a custom agent SDK with warm caches for sub-second cold starts.
Why It Matters
Demonstrates how legacy enterprise SaaS can deploy AI agents at production scale with measurable, team-integrated productivity gains.