A Pragmatic
Introduction to AI

Realistic workflows, common traps, and actual value for product teams.

4 key parts

Intellectual Setup
- Glossary + Cold truths about AI.
Communication & Workflow
- A framework for teams.
Demos
- How to use the tools you already have.
Mental models & Principles
- Why this matters and how I think about AI.

Essential concepts

1. Intellectual Setup

LLM (Large Language Model)

Autocomplete on steroids - predicts next words based on patterns.
Trained on billions of organic + synthetic datasets, but doesn't "understand" - just matches patterns (more on this later).
Why it matters: Explains why are great at writing/coding (patterns) → Bad at math/recent events (needs real knowledge).

Context (Your AI's Memory)

Everything the AI can "see" in your current conversation.
Limited window - like RAM, not hard drive (typically 4k-200k words).
Why it matters: Paste relevant info upfront. AI can't remember what it can't see.

Hallucination (Confident Bullshit)

When AI makes things up but sounds 100% certain.
Happens more with: recent events, specific numbers, obscure topics, URLs.
Why it matters: Always verify facts, dates, numbers, and claims.

Drift (Losing the Plot)

AI gradually forgetting instructions or changing style over long conversations.
Like a game of telephone - quality degrades over time.
Why it matters: For long sessions, periodically remind AI of the task (compacting context).

Token (The Currency)

How AI "sees" text - roughly 1 token = 4 characters.
You pay per token (input + output).
Why it matters: Verbose prompts = expensive. Be clear but concise.

A dose of reality

1. Intellectual Setup

AI Doesn't Understand Anything

Next time you get a response from AI, remember this: It has no idea what it's saying.
It doesn't "know" what a doctor is.
It doesn't "understand" your problems.
It doesn't "comprehend" who or what you are.
It's just INSANELY good at predicting what word should come next.
Why it matters: Start giving better patterns to match. That's what prompting really is.

AI Can't Create Anything "New"

Every single output is a recombination of patterns from training data.
AI has never had an original "thought", and never will.
AI can create combinations that you have never seen before.
Why it matters: AI won't invent new physics laws, but it WILL solve YOUR specific problem in ways you hadn't considered. And that's enough.

AI inherited our biases because it learned from us. If you communicate poorly with humans, you will communicate poorly with AI (the contrary is also true). The principles are the same: Context, Clarity, and Patience."

Core Idea 🔥

A practical guide

2. Communication & Workflow

The missing link

Cooperation = Your part, my part, fingers crossed.
Collaboration = Our solution, shared understanding.
The difference? AI as enabler.

From Cooperation to Collaboration

AI can be the universal translator we've needed.
PMs can create working prototypes → Engineers see the vision, not just requirements.
Engineers can generate UI mockups → Designers better understand technical constraints.
Designers can approach technical documentation → Developers get more accurate implementations.
Everyone can validate* assumptions → No more "I thought you meant...".

The missing link

From Cooperation to Collaboration

Explore

Plan

Execute

Preserve

Clean

Gather

Contextualize

Document

Operate

Reset