Skip to content
AI Fluency

AI Fluency

Tools change. Providers change. The skills do not. Learn the method once and any AI tool gets easier.

Why this exists

Section titled “Why this exists”

At most companies, AI skill is uneven. A few people run whole teams of agents. Many more want to start but never had a free afternoon to learn how it works. That gap is about time, not talent. So people either avoid AI or overpay for it without knowing. That is not their fault.

The fix is to teach the basics, not one product. Nobody learns to code by memorizing a single web framework. They learn the fundamentals, and the next framework is easy. AI works the same way. Learn how it really works, how to steer it, how to check it, and how to pay a fair price, and every tool gets simpler. That includes tools that do not exist yet.

This matters more than it sounds. Prices keep climbing, and people stay on an expensive setup because switching feels like starting over. It is not, once you know the method and not just the buttons. Budgets run dry for a simple reason: nobody mentioned that the biggest, priciest model is rarely the right one for the job. Used well, you do not have to trade quality for cost. Better habits and faster, cheaper models get you more done, sooner, for less.

How to use this guide

Section titled “How to use this guide”

This is not a course you read front to back. Think of it as a row of doors. Pick the one that fits you, read what you need, and come back later.

  • New to AI, or short on time? Start with the five steps below.
  • Using AI for a while and think a course is beneath you? Start with the myths.
  • Here for a specific job? Jump to your role.
  • Want everything? The full chapter list is at the bottom.

If you already build sub-agents, keep a skill library, or run agent teams, most of this will be review. Skim for what is new. If you have time to spare, please help. People who do this well are the ones who keep a guide like this honest. Fixes, better examples, and new chapters are all welcome.

New to AI? Five steps

Section titled “New to AI? Five steps”

These five ideas are the whole method. Know them and you are already ahead of most people. Each one links to a chapter with more depth.

  1. Aim first. Decide what “good” looks like before you type. A vague ask gets confident work on the wrong thing. Say what you want and how you will know it is done. More: don’t prompt, delegate responsibility.
  2. Give the right context, not all of it. The AI only knows what you show it. Too little and it guesses. Too much and the one line that mattered gets lost. More: give it the right context.
  3. Use the cheapest, fastest model that does the job. The top model is rarely the best choice. Match the model to the task and step up only when you have a reason. More: pick the right model.
  4. Check with proof from outside the AI. It sounds sure even when it is wrong. Trust a test, a calculator, or the original source, not its tone. More: what AI really is and catch AI’s mistakes.
  5. Save what works. Write down the brief and the reasoning so the next request takes one line, for you or for a teammate. More: explain it once.

Been at it a while? Test the myths

Section titled “Been at it a while? Test the myths”

If the basics feel too basic, start here. Each belief below is common, wrong, and costs money. Each links to the chapter that takes it apart. If two of them sound familiar, you will get something out of this guide.

  • “The priciest model is the best model.” Often not, and the gap keeps widening. See pick the right model.
  • “AI hardly makes things up anymore.” It still does, with total confidence, especially when answering from memory. See what AI really is.
  • “More context is always better, so paste it all.” Extra text can bury the one line that counted. See give it the right context.
  • “A newer model is always an upgrade.” Not necessarily for your work, and benchmark scores can be gamed. See read model hype.
  • “Small, cheap models are toys.” Many are fast enough to retry again and again, and retrying is how quality happens. See pick the right model.
  • “If we change tools, I start over.” Only if you learned the tool and not the method. See what a harness is.
  • “More agents means faster and better.” Only when the work splits into truly independent pieces. See don’t prompt, delegate responsibility.

Where to start by role

Section titled “Where to start by role”

This section adds no new material. It only tells you what to read first.

Already comfortable with AI. Skip the basics. The value for you is in the tradeoffs.

Leaders.

Engineers, including the skeptics.

  • All the theory chapters, in any order
  • Then the harness chapters, especially sub-agents and skills

Product and research.

Marketing, comms, ops, support, and sales.

All chapters

Section titled “All chapters”

Each chapter takes about 10 minutes and stands on its own. Read them in any order.

Theory: the fundamentals

Harness: the machinery around the model

Appendix: extras

This guide will never be finished. The method outlasts any tool named in it.

Where this came from

Section titled “Where this came from”

I first built this for my friends at Reddit as a hackathon project, to help everyone level up, take the mystery out of these tools, and stop overpaying for AI. It turned out to be useful well beyond one company, so I rewrote it from scratch and put it in the open.

Corrections and new chapters are welcome as pull requests on GitHub. Content is licensed under CC BY 4.0.