Hey man, which one will deploy much faster, API or kid?

his short session by Okcan Yasin Saygili, Devops Lead at Turkcell, explains in a fun way whether Cloud Serverless architecture is really necessary or a trap and what we sacrifice for this architecture.

------------------------------------

Conference Details:
Conference: apidays Australia 2025
Theme: Platforms, Products, and People: The Power of APIs in the Age of AI
Date: 29 - 30 October 2025 • MCEC, Melbourne Australia

--------------------------

Resources from apidays:
Join our upcoming conferences: https://www.apidays.global/
Read the latest API news: https://www.apiscene.io
Explore the API Landscape: https://apilandscape.apiscene.io/

Transcript

1. Hey man, which one will deploy much faster — API or kid? Okcan Yasin Saygili Github: okcan
2. Agenda • 1. Why we compare APIs to kids • 2. The promise of serverless • 3. The hidden trade-offs • 4. What changes when AI joins • 5. Smarter deployment strategies • 6. Real-world example • 7. Key takeaways
3. API vs Kid Compariso n
4. The Serverless Promise Deploy instantly ⚙ ️ No servers to manage Pay per execution Auto-scaling magic
5. The Hidden Complexity ⚠️ Cold starts Cost spikes Vendor lock-in Limited observability
6. What’s Really Fast? Deployment speed ≠ execution speed - Cold starts - Network latency - Model load time - Data transfer overhead
7. When AI Joins the Party Long-running inference Large models (GBs of weights) Stateful data processing Need predictable compute
8. The Cold Start Problem 2–6 seconds startup for AI models Each request reloads weights Breaks real-time applications
9. The Cost Illusion “Serverless is cheap — until it scales.” Model load time = billed runtime Hidden storage and egress fees Observability and logging cost extra
10. The Observabilit y Challenge Scattered logs Fragmented tracing Painful distributed debugging
11. Smarter AI Deployments ✅ Serverless for lightweight logic ✅ Containers for model inference ✅ Hybrid approach (Lambda + K8s) ✅ Predictive autoscaling with ML
12. Real-World Case: AWS Lambda + SageMaker Fraud detection pipeline ⚙ ️ Lambda triggers SageMaker endpoint ⏱️ 3–5s cold start Fixed with scheduled warm-ups
13. Let AI Help the Cloud • Predictive autoscaling • Anomaly-based cost control • ⚡ Workload forecasting • Automatic concurrency tuning
14. Key Takeaways • 1. Fast deploy ≠ sustainable deploy • 2. Serverless hides complexity • 3. AI needs stability • 4. Hybrid architectures win • 5. Use AI to optimize infrastructure
15. Wrap-Up So… which one deploys faster — API or kid? API deploys faster… but the kid learns to adapt better. Like raising a child, good infrastructure grows with time, learning, and care.