Become an AI product engineer

Stage 1 of 9

Start Here

Understand the role, the AI product lifecycle and pick a learning plan that fits your week.

You'll be able to

• Explain what an AI product engineer owns and what they hand off
• Compare the role with AI engineer, ML engineer, PM and full-stack engineer
• Choose a 3, 6 or 12-month plan with realistic weekly hours

Articles in this stage

1. 1.1 What is an AI product engineer? (completed)
2. 1.2 AI product engineer vs AI, ML, product and full-stack roles (completed)
3. 1.3 The AI product lifecycle, from problem to iteration (completed)
4. 1.4 Choose your path: 3, 6 or 12-month plans (completed)

Stage 2 of 9

Product Foundations

Find problems worth solving, write a one-page spec and decide whether AI is the right tool at all.

You'll be able to

• Run problem discovery and user interviews as an engineer
• Write a one-page AI feature spec with success metrics
• Recognize problems where AI is the wrong solution

Articles in this stage

1. 2.1 Problem discovery for AI features (completed)
2. 2.2 User interviews for engineers (completed)
3. 2.3 Jobs-to-be-done for AI products (completed)
4. 2.4 Writing a one-page AI feature spec (completed)
5. 2.5 Defining success metrics before you build (completed)
6. 2.6 When not to use AI (completed)

Stage 3 of 9

LLM Capabilities for Product Decisions

Turn model strengths, latency, cost and context limits into product decisions you can defend.

You'll be able to

• Estimate latency and cost per task before you build
• Choose a model with a repeatable decision framework
• Decide between building, buying and fine-tuning

Articles in this stage

1. 3.1 What LLMs are good and bad at (completed)
2. 3.2 Latency budgets and perceived speed (completed)
3. 3.3 Estimating cost per task (completed)
4. 3.4 Context windows, retrieval and memory trade-offs (completed)
5. 3.5 Choosing a model: a decision framework (completed)
6. 3.6 Build, buy or fine-tune (completed)

Stage 4 of 9

Designing AI UX

Design interfaces that set expectations, show their sources and keep people in control.

You'll be able to

• Design every state of an AI feature, including errors and refusals
• Make outputs editable, undoable and cited
• Add human-in-the-loop approvals where actions are risky

Articles in this stage

1. 4.1 Streaming and loading states (completed)
2. 4.2 Editable outputs and undo (completed)
3. 4.3 Citations and showing sources (completed)
4. 4.4 Communicating confidence and uncertainty (completed)
5. 4.5 Human-in-the-loop and approval flows (completed)
6. 4.6 Error, refusal and empty states (completed)
7. 4.7 Onboarding and setting expectations (completed)

Stage 5 of 9

Rapid Prototyping

Go from idea to a tested demo in days, then turn what you learned into a spec.

You'll be able to

• Validate an AI feature with a prompt-first or Wizard-of-Oz prototype
• Ship a streaming Next.js demo that handles every stop reason
• Turn prototype findings into a buildable spec

Articles in this stage

1. 5.1 Prompt-first prototyping (completed)
2. 5.2 Wizard-of-Oz tests for AI features (completed)
3. 5.3 A day-one demo in Next.js (completed)
4. 5.4 Structured outputs and tool calls in prototypes (completed)
5. 5.5 Turning a prototype into a spec (completed)

Stage 6 of 9

Evals as Product Metrics

Measure quality the way users experience it and catch regressions before they ship.

You'll be able to

• Build a golden dataset from real user tasks
• Design rubrics and calibrate an LLM judge
• Run eval regression suites in CI tied to business KPIs

Articles in this stage

1. 6.1 Why evals are product metrics (completed)
2. 6.2 Building golden datasets from real user tasks (completed)
3. 6.3 Rubrics and human grading (completed)
4. 6.4 LLM-as-judge: design and calibration (completed)
5. 6.5 Regression eval suites in CI (completed)
6. 6.6 Linking eval scores to business KPIs (completed)

Stage 7 of 9

Shipping & Iterating

Roll out gradually, instrument everything and turn user feedback into measurable improvements.

You'll be able to

• Roll out AI features behind flags with staged exposure
• Design A/B tests and analytics events for AI features
• Close the loop from user feedback to prompt changes

Articles in this stage

1. 7.1 Feature flags and staged rollouts for AI (completed)
2. 7.2 A/B testing AI features (completed)
3. 7.3 Product analytics events for AI features (completed)
4. 7.4 Capturing feedback: thumbs, edits and retries (completed)
5. 7.5 Tracing and observability for LLM calls (completed)
6. 7.6 The iteration loop: from feedback to prompt changes (completed)

Stage 8 of 9

Trust, Safety & Economics

Defend against misuse, protect user data and make every AI feature pay for itself.

You'll be able to

• Threat-model prompt injection and tool misuse
• Design input, output and action guardrails
• Model unit economics and price an AI feature

Articles in this stage

1. 8.1 Prompt injection and tool-use risks (completed)
2. 8.2 Data privacy and retention (completed)
3. 8.3 Abuse, misuse and rate limiting (completed)
4. 8.4 Guardrails for inputs, outputs and actions (completed)
5. 8.5 Unit economics of AI features (completed)
6. 8.6 Pricing AI features (completed)
7. 8.7 Cost dashboards and budgets (completed)

Stage 9 of 9

Projects & Career

Turn your projects into a portfolio and case studies, and prepare for AI product interviews.

You'll be able to

• Present projects as product case studies
• Answer AI product sense interview questions
• Plan your next career step and leveling

Articles in this stage

1. 9.1 Building an AI product portfolio (completed)
2. 9.2 Writing a product case study (completed)
3. 9.3 Passing AI product sense interviews (completed)
4. 9.4 Career paths and leveling (completed)