Skip to content

Chapter 0.4 — Demos vs production

🎯 Chapter objective

Show the real distance between an impressive demo and a system in production.

🧠 Core concept

Demos optimize for narrative impact. Production optimizes for stability under real constraints. The two requirements rarely converge.

Dimension Demo Production
Data Carefully chosen Real, dirty, with PII
Latency Tolerant Strict SLA
Cost Marginal Visible and monitored
Errors Anecdotal Formal metric
Operation Manual 24/7 with SLO
Versioning Nonexistent Mandatory
Security Implicit Auditable

🚨 Failure modes

  • Approving an architecture based on a demo.
  • Estimating the production timeline from the demo timeline.
  • Underestimating the cost of continuous operation.

🛡️ Controls and mitigations

  • Define production criteria before starting the project.
  • Treat the demo as a "spike", not an evolvable prototype.
  • Reserve engineering time for the path between demo and production (which is where the real work is).
  • EX-FUND-04 — compare the time, cost and infrastructure of a notebook classifier vs the same classifier exposed via FastAPI with basic observability.

📌 Checklist

  • [ ] Was the demo built with data representative of production?
  • [ ] Are the acceptance criteria for production written down?
  • [ ] Has the operational cost been estimated and approved?