Guides

Twelve guides on the parts of AI product management where most teams get stuck. Eval design, launch gates, agentic products, prompt craft, error analysis.

Pre

Before you vibe code

Use this preflight before opening Cursor, Lovable, Bolt, Codex, or any other AI builder. It should take 10 minutes. If it takes longer, the idea probably isn't clear enough yet. A polished demo will hide that from you.

A week with the playbook

This walkthrough follows Priya, a PM at a mid-stage B2B SaaS company, as she evaluates and specs an AI feature using the playbook artifacts in sequence. The company, product, and people are fictional. The workflow,…

Eval design for PMs

Many generative AI pilots fail to deliver measurable business impact. The most common reason is not bad models or weak prompts. It is missing evals. Teams ship AI features with no definition of "good," no way to detect…

Building agentic products

Agents are one of the defining AI product patterns. An agent is an AI system that takes actions, not just generates text. It reads data, calls APIs, makes decisions, and executes multi-step workflows. Many PMs building…

Operating AI products

AI product management does not end when the model produces a good answer. A vibe-coded prototype can make the AI feel more capable than the product actually is. The real work is deciding when the system can act, who…

Launch gates and staged rollout

The hardest PM judgment call in AI products is not what to build. It is when to say "do not launch." Every AI product has pressure to ship, and the readiness score exists to make that decision based on evidence, not…

Prompt craft for AI PMs

PMs do not need to be the person writing every production prompt. They do need to own the behavioral contract the prompt must satisfy.

Bad to good AI PRD

Most weak AI PRDs do not fail because they are short. They fail because the AI job is vague, quality is undefined, and failure behavior is missing.

Error analysis for AI PMs

Error analysis is the practice of reading real AI interactions, writing down what went wrong, grouping those failures into actionable categories, and using those categories to improve the product.

Artifact flow map

Use this when you need to know which artifact comes next. The playbook is a decision system: each artifact should unlock a specific product decision or make a blocker visible.

Agent PM starter pack

Use this when the AI can call tools, take actions, or run multi-step workflows. Agents need stricter product definition than chatbots because they create side effects.

The AI-native PM loop

When code gets cheaper, product taste matters more. The AI-native PM advantage is not that the PM becomes a full-time engineer. It is that the PM can move from user pain to working prototype to trace review to…