The Roadmap to AI Product Manager

"Be stubborn on vision but flexible on details."
— Jeff Bezos

Source: Photo by Denise Jans on Unsplash

What Makes an AI Product Manager

I frequently get asked by product managers, students, and engineers I work with, “how can I upskill on AI and learn AI product management”. The AI landscape is moving at breakneck speed with new models, frameworks, and capabilities emerging weekly, making it challenging to know where to start or what to prioritize. This guide provides a structured survey of the essential skills needed to excel as an AI PM, organized by domain and prioritized for practical application.

So how does an AI Product Manager differ from a traditional PM? In general, traditional PMs build scalable systems with stable, time-tested tools in a largely deterministic environment. AI product management, on the other hand, often feels like riding a tiger, with the landscape shifting faster than you can map it. Your tools are often weeks old, cutting-edge research is part of the day job, and the products you’re guiding are probabilistic by nature, behaving differently depending on the data they see and often exhibiting jagged edges of performance. Success isn’t measured in features shipped, but in nuanced metrics like model quality, inference efficiency, and real-world generalization. And just when you think you’ve found your footing, a new framework, model, or chip drops, because the open-source community and AI research labs are working 996’s and never stop shipping.

The development process also differs dramatically. Instead of the generally linear Software Development Lifecycle (SDLC), AI systems require a a blend of linear SDLC and highly iterative Model Development Lifecycle (MDLC). This involves continuously tweaking model training pipelines, feature stores, real-time inference latency, versioning of data and models, and monitoring for drift and degradation—all while coordinating with a diverse team of data scientists, ML engineers, infrastructure teams, data SMEs, end users, and model governance stakeholders. This guide maps out the skills you’ll need to navigate this complex, fast-moving landscape successfully.

Skill Sets

Traditional Data Science

A strong foundation in traditional data science is essential for AI PMs because it provides the fundamental language and concepts needed to communicate effectively with data scientists and ML engineers. Understanding probability, statistics, machine learning theory, and evaluation metrics enables PMs to make informed decisions about model selection, set realistic performance expectations, and translate technical capabilities into business value. Without this knowledge, PMs risk making promises that can’t be delivered, misinterpreting model performance, or failing to identify when simpler statistical or deterministic methods might be more appropriate than complex ML or Gen AI solutions.

Probability & Statistics

Fundamental Concepts

Probability distributions (normal, binomial, Poisson, exponential)
Conditional probability and Bayes' theorem
Expected value, variance, and standard deviation
Central Limit Theorem
Law of Large Numbers

Statistical Inference

Hypothesis testing (null hypothesis, p-values, significance levels)
Confidence intervals
Type I and Type II errors
Statistical power
Bayesian vs. frequentist approaches

Descriptive Statistics

Measures of central tendency (mean, median, mode)
Measures of dispersion (range, variance, standard deviation, IQR)
Skewness and kurtosis
Percentiles and quartiles

Correlation and Causation

Understanding correlation coefficients (Pearson, Spearman)
Distinguishing correlation from causation
Confounding variables
Spurious correlations

Sampling Methods

Random sampling
Stratified sampling
Cluster sampling
Bias in sampling
Sample size determination

Machine Learning Theory

Supervised vs. unsupervised learning
Regression, classification, clustering, association, and dimensionality reduction
Familiarity with statistics and ML models and when to use one over the other

Linear Regression
Logistic Regression
Random Forest
Support Vector Machines (SVM)
eXtreme Gradient Boosting (XGBoost)
K-means Clustering
Light Gradient Boosting Machine (LightGBM)
Categorical Boosting (CatBoost)

Model hyperparameters
Grid vs. random search (gradient descent)
Loss functions
Overfitting and underfitting
Bias and variance
Model evaluation techniques (train/test split, cross validation, metrics)

Deep Learning Theory

Model architectures for Artificial Neural Networks (ANN) and deep learning models (DNN, CNN, RNN, etc.)

Activation functions

Rectified Linear Unit (ReLU)

Leaky ReLU
Softplus/SmoothReLU
Parametric ReLU (PReLU)
Exponential Linear Unit (ELU)
Gaussian Error Linear Unit (GELU)

Sigmoid

Loss functions/cost functions

Optimizers

Adadelta
Adafactor
Adaptive Gradient Algorithm (AdaGrad)
Adaptive Moment Estimation (Adam)
Adam with Weight Decay (AdamW)
Adamax
Follow-the-Regularized-Leader (Ftrl)
Lion
LossScaleOptimizer
Nesterov-accelerated Adaptive Moment Estimation (Nadam)
Root Mean Square Propagation (RMSprop)
Stochastic Gradient Descent (SGD)

Backpropagation
Epochs
Learning Rate
Batch Size
Regularization

Early Stopping
Parameter Norm Penalties
- L1 Regularization
- L2 Regularization
- Max-norm Regularization
Dataset Augmentation
Noise Robustness
Sparse Representations

Model weights and biases
Embedding dimensionality: Lower dimensions can lead to less accuracy, more lossy compression. Higher can lead to overfitting and slow training time. A good starting point for the number of embedding dimensions is Dimensions \(\approx \sqrt[4]{\mathrm{Possible\ values}}\)

AI Evaluation Metrics

Theory

Build an intuitive understanding of the right metrics for each model and use case, especially trade-offs.
Recognize areas of concern or blind spots for each metric.
Consider business context when selecting evaluation metrics (e.g., cost of false positives vs. false negatives).

Regression Eval Metrics & Techniques

Mean Absolute Error (MAE)
Mean Squared Error (MSE)
Root Mean Squared Error (RMSE)
Mean Absolute Percentage Error (MAPE)
R-squared (R²) / Coefficient of Determination
Adjusted R-squared
Mean Bias Error (MBE)

Classification Eval Metrics & Techniques

Accuracy
Precision
Recall (Sensitivity)
F1-Score (harmonic mean of precision and recall)
F-beta Score (weighted F1 for precision/recall trade-offs)
Specificity (True Negative Rate)
Confusion Matrix
Precision-Recall Curve
Receiver Operating Characteristic (ROC) Curve
Area Under the ROC Curve (AUC-ROC)
Area Under the Precision-Recall Curve (AUC-PR)
Log Loss (Logarithmic Loss)
Matthews Correlation Coefficient (MCC)

Multi-class Classification Metrics

Macro-averaged metrics (precision, recall, F1)
Micro-averaged metrics
Weighted-averaged metrics
Per-class metrics

Imbalanced Dataset Metrics

When to use precision vs. recall vs. F1-score
Precision-Recall AUC (often better than ROC-AUC for imbalanced data)
Balanced Accuracy
Cohen's Kappa

Clustering Metrics

Silhouette Score
Davies-Bouldin Index
Calinski-Harabasz Index
Inertia (within-cluster sum of squares)
Adjusted Rand Index (ARI) - for labeled data
Normalized Mutual Information (NMI) - for labeled data

Evaluation Techniques

Train/Test Split
K-Fold Cross-Validation
Stratified K-Fold Cross-Validation
Time Series Cross-Validation (for temporal data)
Bootstrap Sampling
Leave-One-Out Cross-Validation (LOOCV)
Holdout Validation

Model Performance Beyond Metrics

Inference latency and throughput
Model size and memory requirements
Training time and computational cost
Robustness to adversarial examples
Fairness and bias metrics
Explainability and interpretability scores

ML & Deep Learning Frameworks

Model Development Lifecycle (MDLC)

Understand the end-to-end MDLC process of building, testing, deploying, monitoring, and decommissioning machine learning models.

Machine Learning Operations (MLOps)

Learn the principles and practices of maintaining and scaling ML workflows in production environments.

ML Pipelines

Pipeline Phases

Data Preparation

Data Extraction
Data Analysis
Data Preparation

Model Development

Model Training
Model Evaluation
Model Validation

Model Serving

Model Registry
Model Prediction
Model Performance Monitoring

Python

Programming paradigms: Functional and Object-oriented programming (OOP)
Data manipulation: Pandas, NumPy, PySpark, Koalas
Package management: uv, pyproject.toml
Documentation: README.md, Agents.md
Testing & Linting: pytest, ruff

SQL

Ensure fluency in querying and manipulating structured data from relational databases.

Distributed Computing & Big Data Processing

Leverage distributed computing for large-scale data processing.
Use PySpark for writing scalable, Python-based ETL and analysis pipelines.

Data Warehouses & Lakehouses

Gen AI & Foundation Models

As generative AI and foundation models become central to modern AI products, PMs must understand their capabilities, limitations, costs, and operational requirements. Knowledge of transformer architectures, scaling laws, and fine-tuning techniques allows PMs to make strategic decisions about when to use pre-trained models versus building custom solutions, estimate compute costs and infrastructure needs, and set appropriate expectations for solution performance with executives. This expertise is crucial for navigating the rapidly evolving landscape of foundation models and ensuring products leverage the latest advances while maintaining cost efficiency and reliability.

Transformer Theory

Architecture & modeling

Attention mechanism
Positional encoding
Tokenization & vector embeddings
Decoder-only, encoder-only, and encoder-decoder architectures
Key-Value (KV) Cache
Hyperparameters

Temperature
Top-K
Top-P

Multi-head Latent Attention (MLA)
Mixture-of-Experts (MoE)

Resources:

Gen AI Scaling Laws

Pre-training Scaling (parameter, data, and training compute size)
Post-training Scaling (RLHF)
Test-time Scaling/Inference Scaling
Multi-Agent Scaling

Diffusion Models

Fundamentals

Forward diffusion process (adding noise to data)
Reverse diffusion process (denoising to generate samples)
Noise schedule and timesteps
Score-based generative models

Architecture & Techniques

Denoising Diffusion Probabilistic Models (DDPM)
Denoising Diffusion Implicit Models (DDIM)
Latent Diffusion Models (LDM)
Stable Diffusion architecture
Classifier-free guidance
Conditional generation (text-to-image, image-to-image)

Applications

Image generation
Text-to-image synthesis
Image inpainting and editing
Super-resolution
Audio generation
Video generation

Training & Optimization

Training objectives (variational lower bound, score matching)
Sampling strategies (DDPM, DDIM, DPM-Solver)
Inference speed optimization
Distillation techniques

Resources

Generative Adversarial Networks (GAN)

Fundamentals

Generator and discriminator networks
Adversarial training process
Minimax game formulation
Nash equilibrium in GAN training
Loss functions (adversarial loss, feature matching)

Architecture Variants

Deep Convolutional GAN (DCGAN)
Progressive GAN (ProGAN)
StyleGAN and StyleGAN2
Conditional GAN (cGAN)
Wasserstein GAN (WGAN)
Least Squares GAN (LSGAN)
CycleGAN (for image-to-image translation)
Pix2Pix

Training Challenges & Solutions

Mode collapse
Training instability
Vanishing gradients
Non-convergence issues
Techniques for stable training (spectral normalization, gradient penalty)

Applications

Image generation
Image-to-image translation
Data augmentation
Super-resolution

Evaluation Metrics

Inception Score (IS)
Fréchet Inception Distance (FID)
Kernel Inception Distance (KID)
Precision and Recall for distributions

Resources

Foundation Model Fine-Tuning

Parameter-Efficient Fine-Tuning (PEFT)

Low-Rank Adaptation (LoRA)
QLoRA (Quantized LoRA)
Adapter modules (Houlsby, Pfeiffer)
Prefix tuning
Prompt tuning
P-tuning / P-tuning v2
IA³ (Infused Adapter by Inhibiting and Amplifying)

Quantization

Quantized Low-Rank Adaptation (QLoRA)
Post-training quantization (PTQ)
Quantization-aware training (QAT)
INT8 / INT4 / binary quantization
GPTQ (GPT Quantization)
AWQ (Activation-aware Weight Quantization)
SmoothQuant
BitsAndBytes

Pruning

Structured vs. unstructured pruning
Magnitude-based pruning
Movement pruning
Wanda (Weight and Activation pruning)
LLM-Pruner and similar methods

Knowledge Distillation

Teacher–student distillation
Response distillation
Feature / representation distillation

Resources

Foundation Model Benchmarks

Fundamentals

What benchmarks measure (knowledge, reasoning, safety, coding, etc.)
Leaderboards vs. arenas vs. LLM-as-judge
Limitations of benchmarks (data contamination, narrow tasks, overfitting)
When to trust benchmarks vs. real-world evaluation

General Model Knowledge & Capability

MMLU (Massive Multitask Language Understanding)
HellaSwag
TruthfulQA
ARC (AI2 Reasoning Challenge)
OpenBookQA
Winogrande
PIQA (Physical IQ)

Reasoning Benchmarks

GSM8K (math word problems)
MATH
HumanEval (code)
Big-Bench (BIG-Bench Hard)
DROP (discrete reasoning)
CommonsenseQA
StrategyQA

Coding Benchmarks

HumanEval
MBPP (Mostly Basic Python Programming)
DS-1000
SWE-bench
CodeContests
MultiPL-E

Safety & Alignment

TruthfulQA (truthfulness)
RealToxicityPrompts
BBH (BIG-Bench Hard) safety subsets
Red-teaming and adversarial benchmarks

Multimodal & Long-Context

MMMU, MMMU-Pro
ChartQA, DocVQA
Long-context benchmarks (Needle in a Haystack, etc.)

Aggregate Suites & Leaderboards

Open LLM Leaderboard (Hugging Face)
LMSys Chatbot Arena
MT-Bench
AlpacaEval
HELM (Holistic Evaluation of Language Models)

Resources

Foundation Model Serving

Cloud Model APIs

Local Serving

Ollama

Foundation Model Evaluation

LLM evaluation metrics

Statistical metrics

LLM-as-judge metrics

Evaluation datasets

Golden dataset
Adversarial dataset
Regression dataset

LLM evaluation tools

Ragas
OpenAI Evals

LLM-as-a-judge techniques

Pairwise comparison
Evaluation by criteria (reference free)
Evaluation by criteria (reference-based)

Foundation Model Observability

Langfuse
LangSmith: a developer platform for inspecting, tracing, and evaluating LLM-powered applications built with LangChain or other orchestration frameworks. It enables fine-grained logging of prompts, model inputs/outputs, tool invocations, and intermediate steps, while supporting automated and manual evaluation workflows for performance, latency, and correctness.

AI Interpretability

Interpretability Methods

Post-hoc Explainability
Intrinsic Interpretability
Mechanistic Interpretability

Anthropic's Interpretability Team Publications

Foundation Model Safety, Security, & Guardrails

Content filtering

GCP Model Armor
Prompt injection

Prompt/Context Engineering

Chain-of-Thought (CoT) Prompting
ReAct
Tree-of-Thoughts (ToT) Prompting
(Article) OpenAI Prompting Guide
(Website) Prompt Engineering Guide by DAIR.AI

Retrieval Augmented Generation (RAG)

RAG has become the dominant pattern for building production LLM applications that need access to private or up-to-date information. PMs with RAG expertise can design systems that effectively combine retrieval and generation, choose appropriate vector databases and embedding strategies, and establish evaluation frameworks that measure both retrieval quality and response relevance. Understanding RAG is essential for building or evaluating AI products that go beyond simple chatbot interfaces to create intelligent systems that can reason over large knowledge bases and provide accurate, contextualized responses.

RAG Fundamentals

Vector embeddings
Chunking
Hybrid retrieval
General resources

(GitHub) RAG Techniques by Nir Diamant

RAG Evaluation

RAG evaluation metrics

Context Precision
Context Recall
Content Entities Recall
Noise Sensitivity
Response Relevance
Faithfulness
Multimodal Faithfulness
Multimodal Relevance

RAG Eval Tools

Vector Databases

Vector search prototyping libraries

Production databases

Agentic RAG

Agentic AI

Agentic AI represents the next evolution of AI systems, moving from single-turn interactions to autonomous agents that can plan, reason, and take actions across multiple steps. PMs need to understand agent frameworks, design patterns, and evaluation methods to build products that can handle complex, multi-step workflows. This knowledge is critical for designing agent architectures that balance autonomy with control, ensuring agents can operate safely and effectively in production environments while delivering on ambitious product visions that require sophisticated reasoning and tool use.

Agent Fundamentals

Resources

Agent Evaluation

Resources

(Paper) Agent-as-a-Judge: Evaluate Agents with Agents

Agent Frameworks

LangChain: open-source framework for building agents
LangGraph: a newer-graph-based library within the LangChain ecosystem designed for creating complex, stateful, and multi-agent workflows with explicit state management and the ability to handle loops and cycles.

(Class) AI Agents in LangGraph by DeepLearning.AI

LlamaIndex: specializes in efficiently indexing and querying large datasets for Retrieval Augmented Generation (RAG) applications

(Class) Building Agentic RAG with LlamaIndex

OpenAI Agent SDK: OpenAI's framework for building agentic AI apps in a lightweight, easy-to-use package with few abstractions. It's a production-ready upgrade of their previous experimentatal framework (Swarm).
Google Agent Development Kit (ADK)

(Code Lab) ADK Crash Course - From Beginner To Expert

Agent Protocols

AI Coding Agents, CLIs, & IDEs

AI Coding Axioms and Practices

AI Coding Context Engineering

AGENTS.md: the standard open-source markdown format for guiding coding agents. AGENTS.md is a README for agents, a dedicated, predictable place to provide the context and instructions to help AI coding agents work on your project.

Agentic Coding Axioms and Ideas

The Dumb Zone: the term was coined by Dex Horthy to describe a critical failure point in AI-assisted software development, where AI agents become unreliable as their context windows fill up. Based on his analysis of 100,000 developer sessions, this zone occurs in the middle 40–60% of a large context window, where model recall degrades and reasoning falters
Tokenmaxxing: a slang term emerging in tech, referring to the practice of maximizing the consumption of AI tokens in a deliberate effort to maximize AI automation and, by extension, productivity. It is increasingly being used by software engineers and companies as a benchmark to measure employee engagement with AI tools and employee productivity

Agentic Design Patterns

General Technical Skills

While AI PMs don’t need to be expert engineers, technical literacy across cloud infrastructure, data pipelines, APIs, and system architecture is essential for making informed product decisions and effectively collaborating with engineering teams. Understanding these technical domains enables PMs to estimate development effort accurately, identify technical risks early, design scalable product architectures, and make trade-off decisions between different technical approaches. This foundation ensures PMs can translate between business requirements and technical implementation, bridging the gap between stakeholders and engineering teams.

Public Cloud Infrastructure

Cloud FinOps

Manage cloud financial operations to maximize efficiency and optimize cost.

Continuous Integration, Testing, and Delivery (CI/CT/CD)

Concepts

Continuous Integration (CI): Devs frequently merge code changes into a central repo, triggering automated builds to find integration issues quickly.
Continuous Testing (CT): Automated tests (unit, integration, etc.) run continuously throughout the pipeline, verifying code quality and preventing bugs from reaching production.
Continuous Delivery (CD): automatically prepares code for release, ensuring it's always in a deployable state, ready to be pushed to production with a single click.

Tools

Jenkins
GitHub Actions
AWS Code Pipeline
GCP Cloud Build/Cloud Deploy
Azure Pipelines

DevOps and Site Reliability Engineering (SRE)

Bridge development and operations to ensure scalable, stable, and reliable systems.
SRE focuses on uptime, latency, monitoring, and incident response with a software engineering mindset.

Data Pipelines

Pipeline Technologies

Apache Airflow

GCP Composer
Amazon Managed Workflows for Apache Airflow (MWAA)
Azure Workflow Orchestration Manager

Apache Beam

GCP Dataflow

AWS Glue
Apache Kafka
Kubeflow Pipelines (KFP)
TensorFlow Extended (TFX)

API and Backend Skills

Develop backends with FastAPI or Flask
Implement REST or GraphQL streaming endpoints for AI services
Design authentication and rate-limiting systems
Build WebSocket implementations for real-time AI interactions

Data Validation

Pydantic

Architecture Concepts

Load Balancing: distributes incoming traffic across multiple servers to improve reliability and performance.
Caching: stores frequently accessed data in memory to reduce latency and database load.
Content Delivery Networks: store static assets across edge servers to reduce latency for users.
Message Queue: decouples components by allowing producers to enqueue messages that consumers process asynchronously.
Publish-Subscribe: enables multiple consumers to receive messages from a shared topic.
API Gateway: acts as a single entry point for client requests and handles routing to backend services.
Circuit Breaker: monitors downstream service calls and stops attempts when failures exceed a defined threshold.
Service Discovery: automatically tracks available service instances so consumers can locate providers.
Sharding: splits large datasets across multiple nodes based on a shard key.
Rate Limiting: controls the number of requests a client can make within a given time window.
Consistent Hashing: distributes data across nodes while minimizing reorganization when nodes are added or removed.
Auto Scaling: automatically adds or removes compute resources based on system metrics.

General Product Management Skills

Core product management skills remain fundamental for AI PMs, but they take on new dimensions when applied to AI products. Project management frameworks help structure the iterative, experimental nature of AI development. Metrics and analytics become more complex when dealing with probabilistic systems where success isn’t binary. Stakeholder management requires explaining uncertain outcomes and managing expectations around model performance. These foundational PM skills are essential for delivering AI products that not only work technically but also create genuine user value and business impact.

Project Management Frameworks

Waterfall: A traditional, sequential approach where each project phase is completed before the next begins. Each phase has specific deliverables and a review process, making it suitable for projects with clearly defined requirements and predictable outcomes. However, it offers limited flexibility for changes once a phase is complete.
Agile: An iterative and incremental approach, suitable for projects with evolving requirements

Scrum: structured roles, sprints, and ceremonies.
Kanban: visual flow-based system emphasizing WIP limits and continuous delivery.

Lean:

Builder Personality Traits

Agency: acts without waiting to be asked. Sees a gap, forms a point of view, and moves. Has a track record of starting things and owning their successful completion.
Relentlessness: the real world rewards the person who gets it wrong nine times and gets the tenth right. AI is making the first 90% of everything trivially easy. The last 10%, the hard, unglamorous, want-to-quit part, is where everything that matters gets built.
Curiosity: not surface-level curiosity where you "wonder why” and move on. The deep kind. The kind where you can’t let something go until you actually understand how it works and could reinvent it from scratch. Where one question leads to another and another and you lose track of time because you need to know.
"Kind Human" Policy: prioritizes kindness, respect, empathy, and professional courtesy at all times, treating these as mandatory, non-negotiable behaviors rather than optional perks. Ensures that performance pressures, high-stakes deadlines, or stressful situations do not result in toxic or disrespectful treatment of others.
Expert Storyteller & Writer: can explain the what, how, why, and when of their product at the drop of a hat. Can explain this equally well to the CEO, the new intern, the lead engineer, or an external client.

Product Metrics

User Adoption & Engagement

Daily/Monthly Active Users (DAU/MAU): the number of unique users who interact with the system daily/monthly
DAU/MAU Stickiness: measures how frequently users return to the product (higher ratios = better)
Feature Adoption Rate: percentage of users who use a specific feature within a given time period

Customer Satisfaction

Net Prompter Score (NPS): measures customer loyalty and satisfaction by asking users the likelihood of product recommendations
Customer Satisfaction Score (CSAT): measures customer happiness with a specific feature or interaction
System Usability Scale (SUS): measures the perceived usability of a product or system (range: 0-100 with higher scores indicating better usability). Importanlty, SUS doesn't measure the satisfaction with the product, so SUS + NPS is a good strategy.
Support Ticket Ratio: number of support tickets per user. A high number can signal product friction, bugs, or adoption issues.

Retention & Churn

Churn Rate: percentage of customers who stop using the product within a given timeframe
Retention Rate: percentage of customers who continue to use the product over a specific period. Retention is important as it's generally more cost efficient than new customer acquisitions

Financial Performance

Customer Acquisition Cost (CAC):
Customer Lifetime Value (CLTV):
Monthly Recurring Revenue (MRR):
Annual Recurring Revenue (ARR):
Average Revenue Per User (ARPU):

Product Performance & Delivery

Robust Documentation

Ensure product documentation is clear, current, and accessible to cross-functional teams.

User Experience (UX) Testing

Test Methods

Wizard of Oz (WOZ) Test: a UX research method where participants interact with a system they believe is autonomous but is actually being operated by a human. It enables testing complex, expensive, or non-existent technology before investing in development

Stakeholder Management

Adept at influencing executives and building consensus in a constantly changing and fast-paced environment.

Product Launch Experience

Know what to do at each product launch stage and how to execute effectively to get things over the finish line

User Journeys

Define clear user journeys aligned with a strategic AI product philosophy and a north star metric.

Time Management and Productivity

Eisenhower Matrix: organizing tasks into four quadrants, divided up by “urgency” on one axis and “importance” on the other
Covey Matrix: uses the same grid as the Eisenhower Matrix, but focuses on long-term planning instead of on daily tasks.
Pomodoro Technique: Pomodoro means “tomato” in Italian, and this method is named for the tomato-shaped timers originally used for it. The technique is simple to follow, set a timer for 25 minutes, work, and then set it for a five-minute break. Repeat the process four times.
Flowtime Technique: Sometimes called the “Flomodoro,” this method tries to help workers break through to a flow state. Users set a plan for the day and divide up their goals into small, manageable tasks. From there, they work without timers, taking breaks as needed. It’s essentially a deconstructed version of the Pomodoro Technique
Eat the Frog: Mark Twain said if you have to eat a frog, do it first thing in the morning. This approach to work is straightforward: Always do the most difficult and dreaded task first. Avoid multitasking and home in on the single task early. Then, the rest of the day will seem breezy by comparison.

Common Product Laws & Axioms

Goodhart's Law: when a measure becomes a target, it ceases to be a good measure
The Mythical Man Month

Brooks's Law: The central thesis that adding more people to a late project increases the delay due to the exponential growth in communication paths and ramp-up time
Man-Month Myth: The idea that a "man-month" is a standard unit of work is deceptive; men and months are not interchangeable because tasks aren't always perfectly divisible

The Lindy Effect: a technology principle that suggests that the future life expectancy of a non-perishable technology, system, or idea is proportional to its current age. Coined by Nassim Taleb, this principle argues that the longer a technology has already survived, the higher the probability it will continue to endure, as it has proven its resilience and robustness over time
Hersey-Blanchard Situational Leadership Model: leaders must adapt their style—ranging from directing to delegating—based on the "readiness" or maturity level of their followers (competence and commitment). Effective leadership requires matching one of four leadership styles (S1-S4) to the follower's development level (R1-R4)

AI Specific Product Skills

AI-specific product skills distinguish exceptional AI PMs from traditional PMs. AI product sense involves understanding when AI is the right solution versus when simpler approaches would be more effective. Experiment design for AI products requires knowledge of A/B testing with probabilistic systems, evaluation metrics that capture model quality, and iterative development cycles that account for model training and refinement. Market insight in the AI space is crucial given the rapid pace of innovation, where new models and capabilities emerge weekly and can fundamentally change what’s possible. These skills enable PMs to navigate the unique challenges of building and launching AI products successfully.

AI Product Sense

Understand what can, and importantly cannot, be solved by AI (i.e. AI is not a silver bullet, many processes and products are better served with non-AI solutions)

AI Experiment Design

Practice iterative hypothesis testing with quantitative evaluation.
Lead with A/B test, user interviews, and user feedback loops where possible

AI Market Insight

Build a deep understanding of the AI market, its competitive landscape, and emerging trends

Newsletters

Tech News

Data Points: a twice‑weekly series from DeepLearning.AI of the most important AI tools, model releases, research findings, and industry developments
Daily Zaps: a daily high-level tech news roundup that trends more business than technical
The Download from MIT Technology Review: a daily high-level tech news roundup from MIT Tech Review
Tech Brew: a punchy, daily roundup of general technology news from the editors of the popular Morning Brew newsletter.

Cloud Developer Programs

Google Developer Program: stay up to date with the latest GCP releases and features
Microsoft.Source newsletter: the curated monthly developer community newsletter provides the latest articles, documentation, and events.
AWS Builder Center: Connect with other builders, share solutions, influence AWS product development, and access useful content.

Engineering Deep Dives

TheSequence: A weekly series that does technical deep dives on the latest AI/ML techniques
The Batch @ DeepLearning.AI: a weekly deep dive from Stanford Professor Andrew Ng
The MLOps Newsletter: technical with a specific focus on MLOps
The Variable: a curated list of articles/tutorials from Towards Data Science
Turing Post: a weekly newsletter by Ksenia Se that covers AI and ML curated summaries of hundreds of industry developments, research insights, and historical context
SwirlAI: MLOps and data engineering focused newsletter with great visualizations
ByteByteGo: a weekly newsletter by Alex Xu that delivers concise system‑design deep dives and foundational tech explainers on complex distributed systems topics like Kubernetes, databases, CI/CD, and API design

Podcasts

People

The “Godfathers of AI”
- Yann LeCun: Chief AI Scientist at Meta and a pioneer in optical character recognition (OCR) and convolutional neural networks (CNN). A Turing Award winner and one of the three “Godfathers of AI”.
- Geoffrey Hinton: University Professor Emeritus at the University of Toronto and former Google Brain lead. Popularizer of backpropagation, AlexNet, and deep learning. 2024 Nobel Prize winner in Physics for machine learning with artificial neural networks. Also well known as a mentor with his former graduate students being Alex Krizhevsky, Ilya Sutskever, Yann LeCun, and many other luminaries. In May 2023, Hinton resigned from Google and started speaking out against the dangers of AI.
- Yoshua Bengio: co-Turing Award winne in 2018 with Yann LeCun and Geoffrey Hinton for his work on deep learning. Bengio is the most-cited computer scientist globally (by both total citations and by h-index), and the most-cited living scientist across all fields (by total citations).
Demis Hassabis: CEO and co-founder of Google DeepMind. He was jointly awarded the Nobel Prize in Chemistry in 2024 for his work on AlphaFold and protein structure prediction.
Mustafa Suleyman: CEO of Microsoft AI and former head of applied AI at Google DeepMind
Andrej Karpathy: Former director of Autopilot at Tesla, co-founder of OpenAI, and prolific AI educator.
Fei-Fei Li: Stanford CS professor, co-director of the Stanford Institute for Human-Centered AI, inventor of ImageNet, and former Chief Scientist of AI/ML at GCP.
Eugene Yan: ML, RecSys, LLMs, and engineering
Andrew Ng: founder of Coursera, DeepLearning.AI, Stanford AI computer science professor, and neural network pioneer

Books

Pop Tech

Narrative and investigative books about major tech companies, case studies, and industry shifts. These works build intuition about how technology intersects with power, markets, and society; helping PMs think strategically beyond the model and anticipate real-world consequences.

Geopolitics

Artificial General Intelligence (AGI)

The Path to AGI: Artificial General Intelligence by John K. Thompson
The Intelligence Explosion: When AI Beats Humans At Everything by James Barrat
Superagency: What Could Possibly Go Right with Our AI Future by Reid Hoffman and Greg Beato

Biographies & Memoirs

The Nvidia Way: Jensen Huang and the Making of a Tech Giant by Tae Kim
Source Code: My Beginnings by Bill Gates

Product Management

Foundational texts on building and scaling software products. They focus on execution, tradeoffs, and organizational dynamics, grounding AI product work in timeless principles for managing complexity and uncertainty.

The Mythical Man-Month: Essays on Software Engineering by Frederick Brooks

Textbooks & Technical Writing

Technical books covering statistics, data analysis, and machine learning fundamentals. They give AI Product Managers the literacy needed to collaborate effectively with technical teams and make informed decisions about model capabilities and limitations.