AI Engineering vs Designing Machine Learning Systems: which should you read?
Short answer
Read AI Engineering first if you build on foundation models such as GPT, Claude or Gemini, and Designing Machine Learning Systems first if you train models on your own data. Chip Huyen wrote them as companions: the 2022 book covers systems built on traditional machine learning, the 2024 book covers applications built on foundation models. Teams that ship AI products in 2026 usually need ideas from both.
By Dr. Dulan Dias, co-founder and CTO of Tellme AI
Key facts
- Designing Machine Learning Systems was published by O'Reilly in 2022 and AI Engineering by O'Reilly in December 2024, both written by Chip Huyen.
- Huyen describes Designing Machine Learning Systems as focused on applications built on traditional ML models, with more tabular data annotation, feature engineering and model training, and AI Engineering as focused on building applications on top of foundation models.
- AI Engineering's repository says it is not a tutorial book, so it does not have many code snippets.
- Both books have free companion repositories with chapter summaries and further resources.
- A June 2025 review on Tensorlabbet found that AI Engineering has hardly any overlap with Designing Machine Learning Systems.
- From Model to Product (independently published, 2026), by this site's author, treats classical ML, computer vision and language models inside one product.
Definitions
- Foundation model
- A large model trained on broad data, such as GPT, Claude or Gemini, that is adapted to a task through prompts, retrieved context or fine-tuning rather than trained from scratch.
- Traditional machine learning system
- A system built around models a team trains on its own labelled data, such as a fraud classifier or a recommender, where features, training data and retraining are the main levers.
- Distribution shift
- A change between the data a model learned from and the data it meets in production, which degrades it silently unless someone monitors for it.
How do the two books differ?
The split is about where the model comes from. If your team trains the model, most of the risk sits in data, features and retraining, which is Designing Machine Learning Systems territory. If you call or adapt someone else's model, most of the risk sits in evaluation, context and cost, which is where AI Engineering spends its pages.
| Designing Machine Learning Systems | AI Engineering | |
|---|---|---|
| Published | O'Reilly, 2022 | O'Reilly, December 2024 |
| Starting point | Your own data and models you train | A foundation model you adapt |
| Core skills | Training data, features, model development, deployment, monitoring, continual learning | Evaluation, prompt engineering, RAG, agents, fine-tuning, dataset engineering, inference optimisation |
| Code | Light; ideas and case studies | Very little, by design |
| Best for | ML engineers and data scientists shipping trained models | Software engineers building LLM features and products |
What Designing Machine Learning Systems covers
The book treats an ML system as a chain of decisions. It starts with when machine learning is the right tool at all and how to frame the problem, then works through data engineering, training data and labelling, feature engineering, model development and offline evaluation, deployment and prediction services, distribution shift and monitoring, continual learning and testing in production, MLOps infrastructure, and the human side of ML.
Its examples are classical models, but the chapters on distribution shift, monitoring and continual learning apply to any model in production, a language model included. That is why it keeps appearing on 2026 reading lists despite its age.
What AI Engineering covers
AI Engineering starts from a capable foundation model and asks how to turn it into a product. It covers how foundation models work at the level an application builder needs, evaluation methodology and evaluating AI systems (including using models to judge models), prompt engineering, RAG and agents, fine-tuning, dataset engineering, inference optimisation, and the architecture and user feedback loops of an AI application.
It deliberately avoids tutorials, so the ideas outlast the APIs. The trade is that readers who learn by building will want a hands-on book alongside it; the comparison with LLM Engineer's Handbook covers that pairing.
Which should you read first?
Whichever you start with, read the evaluation and monitoring chapters closely. They are the parts teams most often skip and most often regret skipping.
| You are | Read first | Why |
|---|---|---|
| A software engineer adding LLM features | AI Engineering | It starts where your work starts: an existing model |
| A data scientist moving models into production | Designing Machine Learning Systems | Deployment, monitoring and retraining are your gaps |
| An ML engineer moving to LLM products | AI Engineering | Evaluation and context replace features and labels |
| A tech lead with both kinds of model | Designing Machine Learning Systems, then AI Engineering | Monitoring and shift ideas carry into the newer book |
| New to machine learning | Neither yet | Learn validation, overfitting and metrics first |
Is Designing Machine Learning Systems still worth reading in 2026?
Yes, with one caveat. Its thinking on training and serving data, distribution shift, monitoring and continual learning has not dated, because those problems do not depend on the model type. The caveat is tooling: named tools and platforms from 2022 have moved on, so read those passages for the reasoning and check current documentation for the products.
AI Engineering has the opposite risk. It is current on foundation models, but the field moves monthly, so treat its specific models and benchmarks as examples and its evaluation and architecture chapters as the part that lasts.
Where both books stop
Because Huyen splits traditional ML and foundation models into two books, neither shows both working inside one product, and neither is a guide to writing the serving code. Real products increasingly mix the two: a classifier or rules decide, and a language model explains or drafts.
From Model to Product covers that middle ground: choosing between rules, classical ML, deep learning and a language model for each part of a problem, then building one product that combines an on-device vision model, rules and a language model, with FastAPI services, containers, monitoring and cost control. It is lighter than AI Engineering on LLM evaluation and fine-tuning and does not cover agents. Disclosure: From Model to Product is written by Dr. Dulan Dias, who also writes this site. It appears where it fits, with what it leaves out, and is not ranked above the other books.
Frequently asked questions
Do I need to read Designing Machine Learning Systems before AI Engineering?
No. Each book stands alone, and Huyen wrote them as companions rather than a sequence. Read AI Engineering first if you build on foundation models. Read Designing Machine Learning Systems first if you train your own models or need to understand monitoring and distribution shift in depth.
Is AI Engineering by Chip Huyen good for beginners?
AI Engineering suits engineers who already build software and want to work with foundation models. It is not a programming tutorial and assumes comfort with basic machine learning ideas such as evaluation metrics. Complete beginners usually get more from a hands-on introduction first, then return to it.
Is Designing Machine Learning Systems outdated now that LLMs dominate?
Its tools are dated in places, but its core is not. Distribution shift, monitoring, continual learning and the gap between training and serving data apply to language model products as much as classical ones. Read it for the reasoning and check current documentation for specific tools.
Does AI Engineering include code examples?
Very few. Its repository says it is not a tutorial book, which keeps it useful as APIs change. Engineers who want to build along the way pair it with a code-heavy book such as LLM Engineer's Handbook or Hands-On Large Language Models.
Which Chip Huyen book is better for MLOps?
Designing Machine Learning Systems. Its later chapters cover deployment, prediction services, distribution shift, monitoring, continual learning and MLOps infrastructure. AI Engineering touches operations through inference optimisation and architecture, but its centre is adapting and evaluating foundation models.
What should I read after both of Chip Huyen's books?
For operating models, Reliable Machine Learning applies site reliability engineering to ML. For the data systems underneath, Designing Data-Intensive Applications. For one product that combines classical ML, vision and a language model end to end, From Model to Product, written by this site's author.
Which Chip Huyen book suits a product manager or founder?
AI Engineering, read selectively. Its opening chapter on whether and how to build on foundation models, and its chapters on evaluation and architecture, give a product lead the vocabulary to judge plans and estimates. Designing Machine Learning Systems matters more if your product depends on models trained on your own data.
How does From Model to Product compare with Chip Huyen's books?
From Model to Product, by this site's author Dr. Dulan Dias, sits between them: one volume that covers rules, classical ML, computer vision and language models in a single product, with serving code, deployment and monitoring. Huyen's books go deeper on their halves, especially LLM evaluation and fine-tuning.
Sources
About the author
Dr. Dulan Dias is co-founder and Chief Technology Officer of Tellme AI in Belfast; founder of EchonLabs, grown from 2017 to more than 60 people before its acquisition in 2026; holder of a PhD in AI and data science from Queen’s University Belfast; and the author of From Model to Product.