monday.com: Internal Agent System Sphera Runs Production AI Agents on Amazon Bedrock with 7 AWS Services

What is Sphera, and why did monday.com build it?

monday.com, a company known for its work management platform, built an internal agent system — a system of autonomous AI agents that carry out tasks without constant human oversight — called Sphera on top of the decade-old codebase of its platform. The system uses seven AWS services — SNS, SQS, EKS, RDS, ElastiCache, EFS, and S3 — as infrastructure, along with Amazon Bedrock, a managed AWS service for accessing large language models, as the inference engine for the agents.

How much have AI tools changed the development cycle?

Before Sphera was introduced, about half of monday.com’s engineers used AI coding tools monthly; today, 9 out of 10 engineers do. This shift in adoption is tracked through a measurable outcome: pull request (PR) throughput, meaning the number of approved and merged code changes per engineer, has grown by more than 50%. Merge rate, the rate of code successfully merging into the main branch without a later revert, remains under control thanks to automated checks before merging.

Morphex and five operational retrofits

An autonomous agent named Morphex, part of the Sphera system, independently proposes and merges code changes with a 95% automatic merge rate and a low percentage of reverts, undoing changes. This reliability rests on five operational “retrofits” the team introduced afterward, while working on the existing system: automated evals, automated evaluations of agent response quality; file-based memory, memory based on files instead of solely on conversation context; remote sandboxes for isolated code execution; automatic PR guardrails, protective checks on pull requests; and monday boards as a shared-state layer, a common source of truth about task status for all agents.