AI Product Management (Aman Khan) (z-library.sk, 1lib.sk, z-lib.sk)

📄 File Format: PDF

💾 File Size: 2.2 MB

📖 Read Online ⬇️ Download

Registered users can read the full content for free

Register as a Gaohf Library member to read the complete e-book online for free and enjoy a better reading experience.

📄 Page 1

(This page has no text content)

📄 Page 2

AI Product Management Build, Evaluate, and Ship Successful AI Products, from Prototype to Production With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles. Aman Khan

📄 Page 3

AI Product Management by Aman Khan Copyright © 2026 AI Product Playbook LLC. All rights reserved. Published by O’Reilly Media, Inc., 141 Stony Circle, Suite 195, Santa Rosa, CA 95401. O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are also available for most titles (https://oreilly.com). For more information, contact our corporate/institutional sales department: 800-998-9938 or corporate@oreilly.com. Acquisitions Editor: David Michelson Development Editor: Angela Rufino Production Editor: Ashley Stussy Interior Designer: David Futato Interior Illustrator: Kate Dullea May 2027: First Edition Revision History for the Early Release 2026-02-23: First Release See https://oreilly.com/catalog/errata.csp?isbn=9798341668270 for release details. The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. AI Product Management, the cover image, and related trade dress are trademarks of O’Reilly Media, Inc. The views expressed in this work are those of the author and do not represent the publisher’s views. While the publisher and the author have

📄 Page 4

used good faith efforts to ensure that the information and instructions contained in this work are accurate, the publisher and the author disclaim all responsibility for errors or omissions, including without limitation responsibility for damages resulting from the use of or reliance on this work. Use of the information and instructions contained in this work is at your own risk. If any code samples or other technology this work contains or describes is subject to open source licenses or the intellectual property rights of others, it is your responsibility to ensure that your use thereof complies with such licenses and/or rights. 979-8-341-66823-2 [FILL IN]

📄 Page 5

Brief Table of Contents (Not Yet Final) Chapter 1: The New Rules of AI Products (unavailable) Chapter 2: Your AI PM Toolkit (unavailable) Chapter 3: Setting Up Your AI PM Playground (available) Chapter 4: Rapid Prototyping with AI (available) Chapter 5: Understanding AI Product Behavior (unavailable) Chapter 6: Developing AI Product Sense Through Teardowns (unavailable) Chapter 7: Practical Building Patterns (unavailable) Chapter 8: Beyond Vibes - Building Evaluation Frameworks (unavailable) Chapter 9: Implementing Evaluations in Practice (unavailable) Chapter 10: The Feedback Loop - Learning from Production (unavailable) Chapter 11: AI Product Strategy (unavailable) Chapter 12: Cross-functional Leadership for AI (unavailable) Chapter 13: From Prototype to Production (unavailable) Chapter 14: Real World Examples (unavailable)

📄 Page 6

Chapter 1. Setting Up Your AI PM Playground A NOTE FOR EARLY RELEASE READERS With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles. This will be the 3rd chapter of the final book. Please note that the GitHub repo will be made active later on. If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at arufino@oreilly.com. “I’m a tool builder. That’s how I think of myself. I want to build really good tools that I know in my gut and my heart will be valuable. And then, whatever happens, is… you can’t really predict exactly what will happen, but you can feel the direction that we’re going. And that’s about as close as you can get. Then you just stand back and get out of the way, and these things take on a life of their own” —Steve Jobs If you could build a working AI application prototype in under an hour, would you still wait three sprints for your engineering team to prototype your idea? This is the central question facing every Product Manager (PM) in the age of generative AI. The old workflow was defined by friction: a critical payment flow bug meant filing a P1 ticket, scheduling a grooming session, explaining the issue across JIRA, and waiting weeks for a fix to be scoped, prioritized, and deployed.

📄 Page 7

One time, I discovered just such a critical bug. Instead of filing a ticket, I took a screenshot, dropped it into Cursor, and asked: “Where in the codebase does this happen? Can you generate a fix?” Twenty minutes later, Cursor had located the bug and generated a fix that my engineering team deployed that afternoon. The first time I did this, it quite literally changed my life by opening me up to the possibilities of what was possible with this technology. Six months ago, that workflow was impossible. Today, it’s widely being adopted daily by technical and “non-technical” PMs alike. The tools enabling this shift - coding agents like Cursor, Claude Code, OpenAI Codex and Factory - give PMs unprecedented power to validate ideas, debug issues, and speak the language of modern AI development. When natural language becomes your programming language, the barrier between idea and implementation disappears. This chapter teaches you to harness these tools to become a more effective PM who can build, test, and iterate without waiting for engineering resources. The goal of this chapter is to reframe treating these coding agents as less of an “execution machine” and more of a technical brainstorming partner. The New Development Stack for Product Managers The bar for technical product management is shifting. Traditional software development follows a predictable pattern: product defines requirements, engineering estimates effort, design creates mockups, engineering builds, quality assurance (QA) tests, and eventually—weeks or months later—users experience the result. This waterfall persists even in Agile teams, hidden beneath two-week sprint cycles and daily standups. AI-assisted development breaks this pattern entirely. Nowadays, with many teams I speak to, the PM is expected to be able to communicate with

📄 Page 8

engineers, designers, and stakeholders across the work much more technically. The toolchain enabling this shift centers on three capabilities: (1) natural language programming, (2) contextual understanding, and (3) iterative refinement. Modern AI development environments (AI IDE) understand intent. Instead of memorizing for loop syntax to write a SQL query, you can prompt, “iterate through each user and find the ones who haven’t logged in this month”. But what does this mean in practice? It means the basic development environment itself has evolved. Let’s examine how the new generation of AI development tools differs fundamentally from the traditional integrated development environments (IDEs) of the past (Figure 1-1).

📄 Page 9

Figure 1-1. An end-to-end development workflow from spec to documentation

📄 Page 10

Understanding Modern AI Development Tools Traditional IDEs are text editors with syntax highlighting and debugging capabilities. They’re powerful but passive - they check whether your syntax is correct, not whether your logic makes sense. AI-powered development environments like Cursor operate differently. They maintain a semantic understanding of your entire codebase. Using AI powered IDEs feels closer to “heads up display” for reviewing and navigating code enhanced by an AI system, more than the traditional “editor” experience (Figure 1-2). Figure 1-2. The coding agent landscape as of December 2024 Here’s an example: when you ask an AI IDE “where does user authentication happen?”, the AI system understands the concept and identifies relevant code even if it uses different terminology like “login flow” or “session management.” This semantic understanding extends beyond reading code. These tools can: Generate entire features from natural language descriptions Refactor code to follow best practices automatically

📄 Page 11

Debug issues from screenshot descriptions alone Write tests that actually catch edge cases Document code in ways humans will actually read For PMs, this means you no longer need to deeply understand syntax to understand systems. You need to understand intent, architecture, and user needs and express these ideas clearly to a system that understands your codebase - exactly the skills you already have. While many powerful AI IDEs exist, we will focus on Cursor as the backbone of our AI PM playground. Cursor: Your AI Technical Partner Cursor reached $100 million in annual1 recurring revenue faster than ChatGPT, becoming the fastest-growing developer tool in history (Figure 1- 3). But thinking of it as just a “developer tool” limits its potential. Cursor functions as an AI technical partner that happens to write code2. The unlock here is really to think about Cursor and any coding agent you use as a technical partner (or tech lead) that never gets annoyed answering your questions.

📄 Page 12

Figure 1-3. Cursor’s growth to $100m ARR compared to other software products To get started here, download Cursor from cursor.sh (select default settings). When you open Cursor, you see four main components that mirror how you already think about products (Figure 1-4):

📄 Page 13

Figure 1-4. The main components of Cursor 1. The Directory Structure (left panel) shows your project’s files and folders—your project’s filing cabinet with everything organized and accessible. Unlike traditional file explorers, Cursor’s directory understands relationships between files. It knows that payment_handler.py relates to checkout_flow.js even though they live in different folders. 2. The Editor (center panel) displays and edits code. Beyond color- coding syntax, Cursor’s editor understands intent. Type “add user authentication,” and it generates the entire authentication flow, including password hashing, session management, and email verification. 3. The Terminal (bottom panel) executes commands—where code becomes reality. When errors occur, Cursor’s terminal explains what went wrong in plain English and often fixes issues automatically.

📄 Page 14

4. The Chat Interface (right panel) connects everything. Ask it questions like “Why is the checkout flow slow?” and get answers referencing specific functions, database queries, and architectural decisions. You’re having a conversation with someone who knows every line of your codebase. Together, these components create an environment where PMs build, debug, and iterate on real applications using natural language. Setting Up Your Environment to Get the Most out of Cursor Let’s go a level deeper into how you can get even more out of your coding agent with the correct instructions and context. First, we need to get our Cursor environment set up. Here’s how to set up Cursor: 1. Learn three shortcuts: – Cmd+T (Mac) or Ctrl+T (Windows): Opens terminal – Cmd+L (Mac) or Ctrl+L (Windows): Opens agent panel – Cmd+K (Mac) or Ctrl+K (Windows): Generates terminal commands with AI 2. Select your model: (recommended): We’ll be spending a lot of time in Cursor Agent - the first thing you should do is turn off “auto” and select the best model you can. This changes monthly - at the time of writing this, the current model stack depends heavily on the task but rotates between the latest OpenAI, Gemini (Google) or Anthropic Claude Model (gemini-3 → claude-4.5-opus —> gpt- 5.2). Use the model with the symbol. The brain means it is a reasoning model (Figure 1-5).

📄 Page 15

Figure 1-5. Selecting a more capable model from Cursor’s options (pick one with a brain icon) If Cursor is blocked by your workplace, sanctioned alternatives include Claude Code CLI, Cline (open source), or Replit. That’s it. You’re ready to build! With your environment installed, the next step is mastering the new interface. Programming with Natural Language The first interaction with the AI sets the tone for everything that follows. After switching to “Ask” mode (a safe space where the AI reads but doesn’t write code), I have PMs type a simple question: “What’s in this codebase? Explain it like I’m a PM, not an engineer.” The shift from syntax to intent represents the most fundamental change in software development since high-level programming languages replaced assembly code. Natural language programming changes how you think

📄 Page 16

about problems. Whereas traditional programming requires decomposing problems into discrete, sequential steps, natural language programming allows us to describe outcomes and delegates decomposition to AI. For instance, building user authentication traditionally requires seven steps: 1. Setting up database tables for users 2. Implementing password hashing algorithms 3. Creating session management logic 4. Building login/logout endpoints 5. Adding password reset functionality 6. Implementing email verification 7. Handling edge cases and errors With natural language programming, you write: “Create a secure user authentication system with email verification and password reset.” The AI generates all seven components, including security best practices you might not know to request. At least, on the surface (more on this later). The Art of Prompt Engineering for Development Effective development prompts share three characteristics: 1. Specificity without prescription: Focus on requirements while letting the AI handle implementation. Say, “create an accessible button that follows iOS design guidelines,” not “create a button that’s 44 pixels tall with 16px padding”. 2. Context awareness: Include context about your domain. “Add user authentication” is different for a banking app versus a social media platform. Include context like, “add user authentication appropriate for a financial application handling sensitive data”.

📄 Page 17

3. Iterative refinement: Start broad, then narrow. Begin with “create a user dashboard,” examine the output, then refine: “add real-time updates to the metrics section” or “make the navigation responsive for mobile devices.” Table 1-1 details a real example from building the AI trip planner. Table 1-1. Prompting examples to improve output Vague prompt Specific prompt Context-rich prompt “Make the background blue” Result: AI changes every blue-ish element on the page, breaking the design “Change only the main content area background to a professional dark blue (#003366) while maintaining all other design elements” Result: Targeted change that preserves the interface “Update the main content area background to match our brand guidelines (dark blue #003366), ensuring proper contrast ratios for accessibility and maintaining visual hierarchy with other page elements” Result: Professional implementation considering accessibility and design system Understanding AI’s Inference Boundaries AI coding assistants excel at inference, but understanding their boundaries helps you provide the right level of detail. AI excels at inferring: Common patterns (user authentication, CRUD operations, API integrations) Security and performance best practices

📄 Page 18

Appropriate error handling for standard scenarios Reasonable UI layouts from high-level descriptions Database schemas from entity descriptions AI needs explicit guidance for: Business logic specific to your domain Custom design requirements unique to your brand Integration details for proprietary systems Performance requirements and constraints Specific user flow preferences Asking AI to “create an e-commerce checkout flow” generates a functional implementation with cart management, payment processing, and order confirmation. But it won’t know your business requires age verification for certain products, offers a loyalty program with specific tier benefits, or needs to integrate with your custom inventory management system. The sweet spot: describe what makes your use case unique while letting AI handle common patterns. Understanding the AI’s boundaries allows you to prompt effectively. Now, we move from theory to implementation. Going From Idea to Working Application with Cursor The traditional path from idea to prototype involves weeks of planning, design, and development. With AI-assisted development, that timeline compresses to hours. For instance, at Arize, a company that implements AI product methodologies in the form of Evals, we use “prototype-first discovery” by building working prototypes in an afternoon, putting them in

📄 Page 19

front of users, and iterating based on real feedback3. Theoretical discussions become concrete experiences. Let me walk you through building your first AI application - an example I love to use is a trip planning assistant that showcases core concepts for any AI product. Step 1: Setting Up Your Development Environment Modern AI IDEs handle infrastructure invisibly. No installing databases, configuring servers, or managing dependencies. First, install Cursor and clone our starter repository. In Cursor, choose “Clone Repository” and enter: https://github.com/amank94/ai-trip-planner This repository contains a foundational trip planner agent. Starting with working code lets you focus on understanding and iteration rather than boilerplate. Once cloned, explore the interface. The file structure shows: trip-planner/ ├── backend/ │ ├── agents/ │ ├── api/ │ └── config/ ├── frontend/ │ ├── components/ │ └── styles/ └── README.md You don’t need to understand every file. The AI handles navigation. Step 2: Your First AI Conversation with Code

📄 Page 20

Before writing code, understand what you’re working with. This is helpful for any new codebase you are working in. Open Cursor’s chat (click the chat icon or press Cmd+L/Ctrl+L). Select a reasoning model (such as in the step previously, i.e. Claude 4.5 Opus or GPT-4.5 from the model dropdown). Switch to “Ask” mode (shift + tab or simply select) and type (Figure 1-6): What's in this codebase? Explain it to me like I'm a PM, not an engineer. Figure 1-6. “Ask mode” in your AI IDE allows you ask questions without the agent generating code. Helpful for learning about a repo or codebase. The AI scans the entire codebase and provides a product-focused explanation: “This is an AI-powered trip planning application that helps users create personalized travel itineraries. From a product perspective: Core Features: Takes user preferences (destination, duration, budget, interests) Generates day-by-day itineraries using AI agents Provides budget breakdowns and local recommendations Exports plans to PDF or calendar formats