{
  "id": 9798215,
  "title": "Your First AI Architecture Project: What Changes and What Stays the Same",
  "url": "https://urgent.news/2026/09/25/your-first-ai-architecture-project-what-changes-and-what-stays-the",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-25T17:29:56.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/tecnovy_academy/your-first-ai-architecture-project-what-changes-and-what-stays-the-same-942"
  },
  "original_language": "en",
  "account": "Your first AI architecture project for an architect involves understanding what remains constant and what new aspects come into play with the integration of AI. In the webinar \"Same Job, New Rules\" by Matthias Bohlen, it was highlighted that while AI changes how the system operates, the architect's role remains crucial. This article outlines the key aspects that stay the same, the changes introduced by AI, and a small-scale project plan to begin with.\n\n**What Stays the Same:**\nAn AI feature remains part of a broader software system, necessitating the consideration of fundamental design questions such as the problem being solved, key quality needs, system boundaries, team ownership, dependency handling, and monitoring. For instance, an AI-assisted incident review system for an operations team still requires setting speed, uptime, privacy, cost, and support goals, defining clear APIs, ensuring safe fallbacks, and documenting design trade-offs.\n\n**What Changes:**\n1. **Variable Results:** AI models can produce varying outputs for the same input, known as probabilistic results. This affects testing methods, requiring tests like output rules, source checks, and safe fallbacks to ensure reliability.\n\n2. **Code vs. Non-Code Behavior:** Unlike traditional services where code and settings primarily dictate behavior, AI systems can also depend on models, prompts, system rules, and search tools. This introduces a need for version control and review for these components, which may change independently of the codebase.\n\n3. **Pipeline Structure:** A simple AI feature might appear as a single box, but the reality is a more complex pipeline involving input sanitization, data retrieval, model call, validation, human review, and logging. Each step can fail independently, making tracing faults challenging. It's crucial to understand whether the AI technology used is Machine Learning (ML) for scoring or classification, Generative AI (GenAI) for content generation, or a combination of both.\n\n**The First Project:**\nThe suggested first AI project for architects should be small, with human review capabilities. For example, an AI-assisted incident review system prepares a draft for engineers without the ability to alter the system or communicate with customers. The project flow is divided into clear segments: input boundary (sanitizing incident data), search (retrieving relevant documentation), model adapter (communicating with the AI model), output checks (validating the AI-generated output), human review (final approval), logs and metrics (monitoring system performance), and a fallback mechanism (retrieving standard search results when the model fails).\n\n**Testing and Prototyping:**\nThe initial prototype should focus on answering design questions rather than solely proving the model's text generation capability. Testing should involve real-world cases to evaluate retrieval accuracy, citation reliability, response times, and costs. Issues like slow model responses, poor search functionality, and private data concerns should be addressed early. A comprehensive test set, including approved and disallowed source files and expected outputs, should be established to validate changes in the AI system.\n\nBy adhering to these guidelines, architects can successfully integrate AI into their projects while maintaining control over the system's reliability, security, and functionality.",
  "summary": "Your first AI project can make architecture work feel old. It is not. The architect still owns the system’s structure, reliability, security, observability, and hard choices. AI does change what drives the system. Some behavior now comes from data, models, prompts, search, and test rules. Code is only one part. This was a key point in Matthias Bohlen’s webinar, “Same Job, New Rules” . AI does not…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}