Most AI agents work in demos and die in production. This framework is about the engineering that closes the gap.
# The 11 Factors
Everything you need to take AI systems from demo to production.
Model Serving Layer
Abstract model providers behind a unified serving layer for portability, failover, and cost optimization.
Context Management
Design and manage the full context pipeline — system prompts, RAG, grounding, and context window strategy.
Memory Management
Give AI systems structured short-term and long-term memory for context-aware interactions.
Integrations (MCP)
Standardise how AI systems connect to external tools, APIs, and data sources.
Orchestration
Coordinate multiple AI components, agents, and workflows for complex task execution.
Human in the Loop
Design clear escalation paths and approval workflows for high-stakes decisions.
Rate Limits & Latency
Handle provider rate limits gracefully and optimise for acceptable response times.
Cost Control
Monitor, budget, and optimise LLM spend with token tracking and model routing.
Evaluation & Observability
Continuously measure, trace, and monitor AI system performance and behaviour.
Safety & Guardrails
Implement input/output filtering, content policies, and behavioural boundaries.
Reproducibility & Audit
Ensure AI behaviour can be replayed, audited, and explained for compliance.
Most AI agents work in demos but fail in production. They hallucinate, they're expensive, they're impossible to debug, and they break in ways nobody predicted. The 11-Factor framework gives you the engineering discipline to build agents that actually ship.
Sandeep Mehta
Founder of OyakoAI. Building multi-agent systems in production — 18 agents on a Mac mini. Previously in healthcare tech. Speaker on AI agents and the gap between demos and real-world deployment.