All paths

馃

The AI Engineering Journey

From early LLM experiments to production agentic systems, the path through 2023-2026.

AI engineering emerged as a distinct discipline somewhere between when ChatGPT launched and when teams started shipping real production systems. This path walks through the formative moments that shaped how AI applications get built today.

9 articles
  1. STEP 1路Sept 2023

    LLM Agents in 2023: Impressive Demos, Harder Reality

    AutoGPT went viral in April 2023. Agents that could autonomously break down goals into tasks and execute them looked like the next leap. Six months later, the gap between demo and production was clear.

  2. STEP 2路Apr 2023

    RAG: The Architecture That Made LLMs Actually Useful for Business

    Retrieval Augmented Generation solved the problem that made LLMs risky for most business applications: they confidently hallucinate facts. RAG grounds responses in real documents and it changed what enterprise AI actually looked like.

  3. STEP 3路Aug 2024

    AI Engineering: The New Discipline Nobody Had a Name For

    Something new emerged in 2024 that sat between software engineering and machine learning but was not quite either. Building reliable products on top of language models required skills and patterns that had not existed before.

  4. STEP 4路Feb 2025

    Agentic AI in Production: What Works in 2025

    Two years after AutoGPT made agents seem imminent, the production reality is clearer. Narrow agents with human oversight work. Fully autonomous agents are still research. Here is the honest picture.

  5. STEP 5路Mar 2025

    RAG Is Harder Than the Diagram Makes It Look

    Every introduction to retrieval-augmented generation shows the same four-step diagram. Building it in production involves considerably more decisions than four.

  6. STEP 6路Jun 2025

    MCP: The Protocol That Made AI Agents Actually Connect to Things

    Anthropic's Model Context Protocol landed in late 2024 and by 2025 had become the standard way to connect AI models to tools and data sources. Here is why it matters and how it works.

  7. STEP 7路Aug 2025

    The Problem With Shipping AI Features

    A team builds a prototype. The prototype is impressive. They ship it. User adoption is lower than expected. The team is confused because the prototype worked well in testing.

  8. STEP 8路Sept 2025

    The Hidden Costs of Production AI

    The cost discussion around AI applications almost always focuses on inference. Once you are at scale, inference is rarely the most important cost.

  9. STEP 9路Feb 2026

    The Enterprise AI Gap

    AI capability has advanced rapidly. The gap between what organisations have access to and what they are successfully deploying is stubbornly wide. The reasons are not technical.

End of path

Browse other paths