Cognitive Intelligence Layer
17 agents, 7 sprints, 23 MCP tools built — zero lines of code changed
Overview
AgentFlow orchestrated 17 specialised agents across 7 sequential sprints to add a complete intelligence layer to the Weaveworks Sock Shop — a live 8-service microservices e-commerce application. Agents built 23 custom MCP tools, generated 26 reports totalling 6,880 lines, and created a fully operational cognitive layer covering revenue analysis, fraud detection, inventory intelligence, customer profiling, and operations assessment. Zero lines of application code were modified.
Methodology
7-sprint sequential execution with parallel agent assignment within each sprint. Phase 1: Deep-dive codebase reconnaissance across all 8 services. Phase 2: Live API probing with 5 parallel agents. Phase 3: Cognitive intelligence layer with parallel revenue, fraud, inventory, customer, and operations agents. Phase 4: Executive synthesis and demo narration.
Results
17 agents completed all 7 sprints. 23 MCP tools auto-generated across 7 domains (catalogue, cart, orders, auth, users, addresses, cards). 26 detailed reports produced. The 'Holy Sock Paradox' was discovered — a product mathematically impossible to purchase ($99.99 + $4.99 shipping exceeds the $100 payment threshold) — found by correlating data across 3 agents.
Key Benefits
Complete intelligence layer without any code modification to the target application
Autonomous MCP tool generation — agents built their own typed API layer
Cross-agent discovery — findings that no single agent could produce alone
Reference architecture for B2C and B2B intelligence layers on any application
See the Evidence
Every output is public. Inspect the code, the reports, and the results yourself.
See More Demonstrations
Explore other real-world demonstrations or get in touch to discuss your own use case.