AI at work — your questions answered

Almost certainly not in the near term — but it will change what "doing your job well" means.

Machine learning is a broad category: systems that learn from data rather than following explicit rules.

It depends on the task.

A large language model is trained on enormous amounts of text to predict the most statistically likely continuation of any given input.

The context window is how much text an AI model can "hold in mind" at once — including your prompt, the conversation history, and any documents you've pasted.

The highest-leverage use cases for PMs: synthesising user research and NPS verbatims into themes, generating and stress-testing experiment hypotheses, writing first drafts of briefs and specs, creating copy variants for A/B tests, building quick prototypes for stakeholder alignment (vibe coding), and automating weekly reporting workflows.

Use the 5-layer framework: Role (who the AI should be), Context (your situation and constraints), Task (the specific output you want), Constraints (format, length, tone, what to avoid), and Examples (1–2 examples of ideal output).

For the free and standard tiers of most AI tools — approach with caution.

Three changes fix 90% of generic output: (1) Add role framing — "You are a senior growth PM at a B2B SaaS company" before every substantive task.

Use a tool like Granola that auto-generates structured meeting notes in the background — so you stay present instead of taking notes.

Vibe coding is building functional software entirely through natural language — describing what you want, iterating on the output in plain English, never directly writing code.

A regular chat starts blank every time — no memory, no standing context.

MCP (Model Context Protocol) is an open standard that lets AI models connect to your internal tools — Notion, Linear, Contentful, your analytics API — in a standardised way.

All three are automation tools that connect apps and trigger workflows.

Yes — but specifically for building prototypes and PoCs, not for writing production code.

RAG stands for Retrieval-Augmented Generation.

Don't pitch AI in the abstract — show a specific before/after.

Three sources, one hour per week: (1) Lenny's Newsletter for PM-specific AI applications, (2) Superhuman AI newsletter for the weekly tool landscape, (3) Actually using AI for one new task each week — hands-on experience compounds faster than reading about it.

In order of leverage: (1) Prompt engineering — the core skill that multiplies everything else.

Be specific, not aspirational.

Yes — specifically if you use AI to skip the thinking rather than to accelerate it.

It's real, with an important caveat.

Three ways: (1) Use AI to generate 10 copy variants instead of 2 — then pick the 2 most differentiated to test.

Four high-value use cases: (1) Copy variants — generate 6 variants of any UI string across different tones, pick the best 2 to test.

The test is whether AI solves a real job-to-be-done better than the non-AI alternative — not whether it's impressive.

Three underrated things: (1) Prompt libraries compound — a team using the same high-quality prompt template consistently outperforms a team improvising.