Reactiv deploys three-agent scheduler for Shopify mobile apps

Reactiv has put a three-agent scheduler into production to generate proposed updates for Shopify merchants’ native mobile apps from recurring natural-language instructions, according to Reactiv and AWS. The system moves work that previously required merchants to edit apps interactively into scheduled runs, but it does not publish changes on its own: proposed configurations are stored for merchant review, and nothing goes live without explicit approval.

A merchant creates a recurring task through a chatbot or form, specifying a prompt, frequency and time. An Amazon EventBridge cron rule triggers an AWS Lambda executor, which validates the merchant account, obtains a concurrency lock, retrieves the current app configuration and passes that context to Amazon Bedrock AgentCore. AWS documents cron-based scheduled rules as a legacy EventBridge feature and recommends EventBridge Scheduler for new scheduled workloads.

Inside AgentCore, a Supervisor Agent classifies the request and routes it to an Analytics Agent, a Builder Agent or both. The Analytics Agent turns text requests into SQL queries against Amazon Redshift. The Builder Agent uses more than 50 tools to create an updated configuration, while a Model Context Protocol server checks each mutation against Reactiv’s schema. The completed proposal is written to Amazon DynamoDB for review.

Reactiv also migrated its interactive dashboard agent to the same Strands-based stack. AWS and Reactiv say the interactive and scheduled paths now share MCP servers and AgentCore Memory, so merchant preferences and facts captured in either mode can inform later sessions. AWS documentation confirms that AgentCore isolates runtime sessions and can organize long-term memory by strategy, actor, session and namespace, but it does not independently verify Reactiv’s tenant mapping or production security setup.

The performance figures come from Reactiv’s internal measurements reported in the AWS/Reactiv case study and were not independently validated. The companies claim merchant configuration time fell 80%, with onboarding work declining from 17 hours to 3 hours and post-launch changes taking minutes rather than days. They also say a three-person team shipped the three-agent system in 10 weeks, versus 15 weeks for a previous single-agent build, which they characterize as 33% faster. They claim native streaming reduced scheduled-job runtime from more than 10 minutes to about 5 minutes, or twice as fast, and that removing custom infrastructure saves nearly $6,000 a year in compute costs. The case study does not disclose the measurement methodology, sample size, production launch date, participating merchant count or task volume.