Architecting AI-Native Systems: A Practitioner's Decision Guide

Authors

  • Ankit Jain

Keywords:

AI-native architecture, model selection, embedding models, memory architecture, LangChain, LangGraph, RAG, prompt engineering, fine-tuning, cloud strategy, decision framework, software architecture, reference implementation, AWS Bedrock, Microsoft Foundry, Google Vertex AI, Databricks.

Abstract

Understanding that system design has entered an AI-native era is necessary but insufficient. Engineering leaders and architects face a cascade of consequential decisions before a single line of agentic code can be written: which capability on the AI spectrum is actually needed, which foundation model and embedding model should ground the system, which implementation stack balances developer control against operational overhead, and how cloud and infrastructure strategy should be adapted for AI workloads. Poor decisions at any of these layers produce systems that are over-engineered, under-performing, or impossible to govern and maintain. This paper provides a structured, practitioner-oriented decision guide for each of these layers. We introduce the Minimum Necessary Capability principle for navigating the prompt-engineering-to-autonomous-agents spectrum, a model selection framework covering frontier LLMs, mid-tier LLMs, and small language models (SLMs) across commercial and open-weight providers, criteria for embedding model selection including domain specificity and dimension-cost trade-offs, a decision framework for implementation stack selection ranging from no-code platforms to developer-controlled frameworks such as LangChain and LangGraph, and a cloud strategy framework covering single-cloud, multi-cloud, and hybrid approaches for AI-native workloads. Memory architecture is treated as a cross-cutting design concern woven into model selection, implementation stack, and the decision playbook rather than as a standalone decision. A deep reference implementation on AWS maps these abstract decisions onto a concrete cloud service architecture, and a cross-platform comparison to Microsoft Azure and Google Cloud demonstrates provider portability of the four-layer framework. An integrated seven-step decision playbook, a worked greenfield example, a migration playbook, and an anti-pattern catalog translate these frameworks into immediately actionable guidance. Together with its companion paper [1], this work provides architects and engineering leaders with both the conceptual foundation and the practical decision tools needed to design, build, and govern AI-native systems that endure.

 

Downloads

Published

2026-09-28

How to Cite

Jain , A. (2026). Architecting AI-Native Systems: A Practitioner’s Decision Guide . International Journal of Artificial Intelligence and Machine Learning, 6(12s), 1702–1722. Retrieved from https://www.svedbergopen.com/index.php/ijaiml/article/view/3180