Reactiv cuts merchant setup time 80% with an autonomous AI Scheduler on Amazon Bedrock AgentCore

# Problem statement Reactiv builds native iOS and Android apps for Shopify merchants. Merchants needed frequent, timely updates to homepages and promotional content because stale app content costs conversions. Before the AI Scheduler, merchants either chatted with an interactive assistant to make changes or handled updates manually. Merchants asked for autonomous, scheduled updates that run without manual intervention.

# What Reactiv built Reactiv created an AI Scheduler that runs on a schedule and performs end-to-end updates. A merchant can give a natural-language instruction (for example, "Refresh my homepage with best sellers every Monday at 9 AM"), and the system executes the query, selects products, updates layouts and assets, validates the changes, and publishes the app update.

They implemented the Scheduler as a three-agent system using the Strands Agents SDK on Amazon Bedrock AgentCore. The three agent roles:

• Supervisor (intent classifier) to interpret schedule requests.
• Analytics agent to query merchant data (sales, top sellers, categories).
• Builder agent to generate and mutate app configurations against a live schema.

They consolidated interactive and scheduled workflows onto one stack to share memory and preferences between live conversations and scheduled jobs.

# Why AgentCore was chosen Reactiv needed four capabilities that their original architecture lacked: multi-agent orchestration, persistent cross-session memory, native Model Context Protocol (MCP) support for their configuration schema server, and lower tool-definition overhead. AgentCore provided:

• Managed agent runtime: Agents run inside Firecracker microVMs. Reactiv packages its Strands agent graph as a Docker image, pushes it to Amazon ECR, and deploys to AgentCore. Agents spin up when schedules trigger and shut down when done, avoiding manual container management and cluster configuration.
• Built-in memory: AgentCore memory persists context across sessions. Reactiv uses three memory strategies: session summarizers, a preference learner for approved/rejected layouts, and a semantic fact extractor for store-specific info (product categories, top sellers, brand guidelines). Memory is scoped per merchant.
• AgentCore. The Builder Agent mutates the live configuration state and validates changes against the schema on every call. AgentCore Identity replaces Reactiv's prior custom Cognito layer and JSON-RPC handshakes, reducing authentication and integration code.
• Multi-tenant isolation: Per-merchant memory and AgentCore runtime isolation keep each merchant's data private without custom routing or vector DB setup.

# Operational results cited by Reactiv

• Merchant configuration time reduced by 80%.
• Time to production shortened by 33%.

# Implementation notes Reactiv packages the agent graph as a Docker image deployed to Amazon ECR, then runs it on AgentCore. Agents are ephemeral, launching on schedule triggers and shutting down when complete. Memory persists across those ephemeral sessions. The Config MCP runs statefully in the AgentCore runtime so the Builder Agent operates on live app state with schema validation.

# Practical implications for similar teams

# Bottom line Reactiv automated recurring app updates for Shopify merchants by combining a small graph of specialized agents with AgentCore's managed runtime, memory, and identity services. The result: faster deployment, less configuration overhead, and shared memory across interactive and scheduled experiences.