Manning Publications Co.

Agentic AI gets much more interesting once it has to interact with actual tools, workflows, and software environments. Not just answer prompts in a sandbox. Building Agentic Applications with CrewAI and MCP by Max Gfeller is a practical guide to designing agentic systems that do real work. You'll start by building a single agent with prompts, tools, and structured outputs, then scale into multi-agent workflows for content creation and documentation. From there, the book moves into MCP servers, Cursor integrations, multimodal systems, CopilotKit-powered chat experiences, and human-in-the-loop workflows. If you're trying to understand how agentic AI starts to resemble production software and not just demos, this is a strong place to start. New in MEAP and half off through May 27th: https://hubs.la/Q04gcL210

RLHF is behind many of the biggest advances in modern AI, but most explanations stay surface-level. Reinforcement Learning from Human Feedback by Nathan Lambert breaks it down from first principles through real training workflows. You'll start with the core papers and concepts, then move into how RLHF actually works in practice — covering optimization methods, evaluation, and techniques like constitutional AI and synthetic data. It also doesn't dodge the hard parts: open research questions, tradeoffs, and where the field is still evolving. The result is a clearer view of how today's models are shaped and where they're headed. Check it out: https://hubs.la/Q04dX0400

Most deep learning optimization happens a layer above the GPU. CUDA forces you to go deeper. And that's where the real gains are. CUDA for Deep Learning by Elliot Arledge shows how to work at the level where performance decisions actually happen. You'll learn how to profile with Nsight Compute, isolate bottlenecks, and understand why optimizations work — not just apply them. By working across multiple levels of abstraction, you build both intuition and hands-on skill: from kernel design to GPU-level debugging. The result is a kind of performance literacy that carries forward, even as hardware evolves. Learn more: https://hubs.la/Q04dWm1K0

What does a focused "C#/.NET learning sprint" actually look like? Julio Quezada put together a five-book reading path designed to go beyond features and syntax. The goal: build better systems thinking — clear separation of concerns, lightweight dependencies, and systems that remain reliable under stress. Two Manning titles made the list: • ASP. Net Core in Action, Third Edition by Andrew Lock: https://hubs.la/Q04d-2db0 • Dependency Injection Principles, Practices, and Patterns by Steven van Deursen and Mark Seemann: https://hubs.la/Q04d-4Cy0 It's a practical mix of hands-on development and architectural clarity, useful whether you're refining your fundamentals or leveling up how you design systems. See the full list: https://hubs.la/Q04d-19F0

Ever wanted to understand deep learning without switching out of R? Deep Learning with R, Third Edition by François Chollet and Tomasz Kalinowski builds that understanding step by step — starting from fundamentals and layering in practical examples using Keras. It's designed to help you not just use models, but actually understand how they work. The third edition expands into what's shaping modern AI: transformers, GPT-style models, and diffusion-based image generation. The result is a guide that stays grounded in code while keeping pace with how the field is evolving. Longtime reviewer Hefin Rhys highlights how the newer material, especially around NLP and transformers, adds meaningful depth, even if you've read earlier editions. Read his full review: https://hubs.la/Q04fZzGJ0 Explore the book: https://hubs.la/Q04fZnZ50

Most LLMs are built to do a little bit of everything. And that's often the problem. If you're working toward specific business goals, general-purpose models can be inefficient, expensive, and misaligned. Rearchitecting LLMs by Pere Martra focuses on reshaping models to fit your domain, not the other way around. You'll move beyond theory and into practical techniques like fine-tuning, pruning, and knowledge distillation to build smaller, more efficient models. The book walks through how to analyze model behavior, remove unnecessary components, and even apply "fair pruning" to address bias at the neuron level. The result: domain-specific SLMs that are faster, cheaper, and easier to deploy without sacrificing performance. Learn more: https://hubs.la/Q04dYp2R0

Modern web apps don't have to be heavy to be powerful. With today's browsers, a lot of what we reach for frameworks to do is already built in. And in many cases, skipping the extra layers means faster load times, fewer security concerns, and simpler systems to maintain. Vanilla Web by Maximiliano Firtman shows what that looks like in practice. You'll build reusable UI components, set up routing, create an installable PWA, and even develop a full e-commerce frontend — all using plain JavaScript, HTML, and CSS. It's not about rejecting frameworks outright. It's about understanding the platform well enough to choose when you actually need them and when you don’t. Explore it here: https://hubs.la/Q04dZ_Bm0

Building a real learning path in AI is harder than it should be. Too many options, and not enough signal. That's why outside perspectives can be useful. Especially from people actively working in the space. Hadi Aghazadeh, author of Applied Reinforcement Learning, points to two books that helped shape his foundation for navigating modern GenAI challenges: • Build a Reasoning Model (From Scratch) by Sebastian Raschka • Reinforcement Learning from Human Feedback by Nathan Lambert Together, they connect how reasoning models are built with how they're aligned — two pieces that increasingly go hand-in-hand. Read his full take: https://hubs.la/Q04dX5D80 Links to all three books are in the comments.

Most teams don't plan for performance — they debug it later. That approach gets expensive fast. In Performance Engineering in Practice, Den Odell lays out a different model: build performance in from the start. His "Fast by Default" approach gives you a structured way to think about speed early before issues compound. The System Paths framework adds a shared diagnostic language your team can use across stacks, making performance easier to reason about and act on. Learn more: https://hubs.la/Q04dWR9N0