Every Python developer we talk to is somewhere on one of three roads right now. Building AI agents. Training models. Or getting the fundamentals solid enough to do either.
This month we are giving away books for all three. Five winners get PDF and ePub copies of our three bestselling Python titles. You can enter in about 30 seconds at packt.link/draw, and everything about the giveaway is at the bottom of this post.
But first, the useful part. Here is what each of those three paths actually looks like in 2026, what you should learn on each, and where to start.
Path 1: Building an MCP app
Model Context Protocol went from announcement to industry standard in about a year. If you have used Claude with tools, VS Code agents, or almost any serious agentic product recently, you have used MCP without thinking about it. It is the layer that lets an LLM call your code, read your data, and act on your systems in a standardized way.
The core mental model is small. An MCP server exposes capabilities. Those capabilities come in three flavors: tools (functions the model can call), resources (data the model can read), and prompts (templates the model can use). An MCP client, like Claude or a VS Code agent, connects to your server and negotiates what is available.
A minimal server in Python looks like this:
from mcp.server.fastmcp import FastMCP
mcp = FastMCP("orders")
@mcp.tool()
def order_status(order_id: str) -> str:
"""Look up the status of an order."""
return lookup(order_id)
if __name__ == "__main__":
mcp.run()
That decorator is doing a lot of quiet work. It generates the schema the model reads to understand what your function does, which is why the docstring and type hints matter more here than anywhere else in your codebase. The model literally reads them.
Where it gets interesting is everything after hello world: managing resources at scale, wiring servers into real clients, describing capabilities so different components interoperate cleanly, security, and cloud deployment. That is the arc Learn Model Context Protocol with Python by Christoffer Noring follows, built around one practical project that grows chapter by chapter until you are shipping agentic apps that work with Claude for desktop and VS Code.
If your 2026 roadmap includes the phrase "AI agent" anywhere, this is the path.
Path 2: Training your next ML model
The by-example route into machine learning still works better than the theory-first route for most working developers. You learn logistic regression properly the day you use it to predict something you care about.
The modern version of this path has a distinct shape. You start with the classics because they still solve most business problems:
from sklearn.ensemble import RandomForestClassifier
model = RandomForestClassifier(n_estimators=200)
model.fit(X_train, y_train)
print(model.score(X_test, y_test))
Then you climb. Tree ensembles to neural networks. Neural networks to transformers, where BERT and GPT stop being buzzwords and become architectures you can fine-tune. Then multimodal, where text and vision models meet, and you build things like an image search engine that understands natural language queries.
The projects are what make the path stick. Predicting stock prices teaches you time series and the humility that comes with them. Building an image search engine teaches you embeddings better than any diagram. Python Machine Learning By Example, Fourth Edition by Yuxi (Hayden) Liu, an ex-Google ML engineer, is structured exactly this way: worked examples first, with two new chapters on NLP transformers with BERT and GPT, and multimodal computer vision with PyTorch and Hugging Face.
If you can already write solid Python and want the ML rungs above it, this is the path.
Path 3: Something simple with Python
The unglamorous truth: both paths above quietly assume your fundamentals are solid. Decorators show up in your first MCP server. Comprehensions, iterators, and OOP are everywhere in ML code. When those feel shaky, every tutorial takes three times longer than it should.
There is no shame in taking this path, whether you are starting from zero or filling gaps you have coded around for years. The problem was never that fundamentals are hard. It is that most fundamentals books are boring enough to be a controlled substance.
Python Illustrated by Maaike van Putten and Imke van Putten takes the opposite bet. It teaches Python from your first function through object-oriented programming with illustrations, analogies, and exercises, narrated by a know-it-all cat with a slightly moody dachshund learning alongside you as a study buddy. It sounds like a gimmick. It reads like the book you wish you had started with. Visual learners especially tend to click with it.
If the honest answer to "what are you building next?" is "the foundation," this is the path.
Why choose? The giveaway
All three books are in our July giveaway.
- 5 winners
- Each winner gets PDF and ePub copies of all three books
- One entry per person
- Entries close July 31
Enter here: packt.link/draw
That is the whole process. No purchase, no hoops, about 30 seconds of your time.
And since the question is the whole point of this post: what are you building next? Tell us in the comments. An MCP server for your team, a model you have been putting off training, or your first real Python project all count.
Top comments (0)