Reactiv used Amazon Bedrock AgentCore to automate scheduled updates to Shopify merchants’ native mobile apps, cutting merchant configuration time by 80 percent and reaching production 33 percent faster, according to its internal measurements. Reactiv says shoppers convert in its apps at 2–4 times the rate of web visitors. Its AI Scheduler lets merchants request updates in natural language, such as refreshing a homepage with best sellers every Monday at 9 AM, then prepares changes for merchant review. Nothing goes live without explicit approval.

The scheduler uses three specialized agents built with the Strands Agents SDK: a Supervisor Agent routes each request, an Analytics Agent queries merchant data, and a Builder Agent creates updated app configurations. AgentCore runs the agents in isolated Firecracker microVMs, stores long-term memory across sessions, and hosts Reactiv’s configuration server through the Model Context Protocol (MCP), a standard for connecting agents to tools. The memory system summarizes prior jobs, tracks approved and rejected layouts, and stores store facts such as product categories, top sellers, and brand guidelines. Memory and execution are isolated for each merchant.

When a merchant’s EventBridge schedule triggers, an AWS Lambda job executor checks the account, locks concurrent work, and retrieves the current app configuration. It sends the prompt and context to AgentCore, where the agents can use more than 50 tools, including MCP configuration changes, Lambda data queries, Shopify product lookups, and image generation. The Analytics Agent runs text-to-SQL queries against Amazon Redshift; the Config MCP validates each change against Reactiv’s app schema. Results go to Amazon DynamoDB for review. Amazon Bedrock provides the agents’ foundation models and guardrails.

Reactiv moved its interactive dashboard agent onto the same Strands framework, AgentCore-hosted MCP servers, and memory instance as the scheduler. The shared memory lets preferences and facts from an interactive session inform a later scheduled job, and scheduled-job learnings inform the dashboard agent. The migration replaced approximately 100 OpenAPI specification files and custom authentication and JSON-RPC handshake code. Reactiv reports nearly $6,000 in annual compute savings and job times reduced from over 10 minutes to approximately 5 minutes, or 2x faster. Onboarding work that required 17 hours of manual effort now takes 3 hours; later changes can be made in minutes instead of days.

A three-person team delivered the three-agent system in 10 weeks, compared with 15 weeks for its earlier single-agent build. Reactiv plans to expose colors, typography, spacing, and component variants through a design-system MCP server, so merchants can request brand-matched changes within defined design properties. It also plans an onboarding flow that analyzes a merchant’s brand name and website to create a starting mobile app, followed by support for merchant-requested layout sections represented as validated JSON and rendered with Reactiv components. Reactiv says its production system now runs three agents, more than 50 tools, and persistent memory shared across scheduled and interactive use.

Reactiv used Amazon Bedrock AgentCore to automate scheduled updates to Shopify merchants’ native mobile apps, cutting merchant configuration time by 80 percent and reaching production 33 percent faster, according to its internal measurements. Reactiv says shoppers convert in its apps at 2–4 times the rate of web visitors. Its AI Scheduler lets merchants request updates in natural language, such as refreshing a homepage with best sellers every Monday at 9 AM, then prepares changes for merchant review. Nothing goes live without explicit approval.

The scheduler uses three specialized agents built with the Strands Agents SDK: a Supervisor Agent routes each request, an Analytics Agent queries merchant data, and a Builder Agent creates updated app configurations. AgentCore runs the agents in isolated Firecracker microVMs, stores long-term memory across sessions, and hosts Reactiv’s configuration server through the Model Context Protocol (MCP), a standard for connecting agents to tools. The memory system summarizes prior jobs, tracks approved and rejected layouts, and stores store facts such as product categories, top sellers, and brand guidelines. Memory and execution are isolated for each merchant.

When a merchant’s EventBridge schedule triggers, an AWS Lambda job executor checks the account, locks concurrent work, and retrieves the current app configuration. It sends the prompt and context to AgentCore, where the agents can use more than 50 tools, including MCP configuration changes, Lambda data queries, Shopify product lookups, and image generation. The Analytics Agent runs text-to-SQL queries against Amazon Redshift; the Config MCP validates each change against Reactiv’s app schema. Results go to Amazon DynamoDB for review. Amazon Bedrock provides the agents’ foundation models and guardrails.

Reactiv moved its interactive dashboard agent onto the same Strands framework, AgentCore-hosted MCP servers, and memory instance as the scheduler. The shared memory lets preferences and facts from an interactive session inform a later scheduled job, and scheduled-job learnings inform the dashboard agent. The migration replaced approximately 100 OpenAPI specification files and custom authentication and JSON-RPC handshake code. Reactiv reports nearly $6,000 in annual compute savings and job times reduced from over 10 minutes to approximately 5 minutes, or 2x faster. Onboarding work that required 17 hours of manual effort now takes 3 hours; later changes can be made in minutes instead of days.

A three-person team delivered the three-agent system in 10 weeks, compared with 15 weeks for its earlier single-agent build. Reactiv plans to expose colors, typography, spacing, and component variants through a design-system MCP server, so merchants can request brand-matched changes within defined design properties. It also plans an onboarding flow that analyzes a merchant’s brand name and website to create a starting mobile app, followed by support for merchant-requested layout sections represented as validated JSON and rendered with Reactiv components. Reactiv says its production system now runs three agents, more than 50 tools, and persistent memory shared across scheduled and interactive use.