You Don't Need to Believe in AI.
You Need to Test It.

We're not going to start by telling you what AI is. We're going to start by having you use it, right now, and form your own judgment.

Now you have something more valuable than any briefing could give you: your own data point. Whatever you rated that answer, you now know something about what AI can do from direct experience. Everything that follows will help you understand why you got the result you did — and how to get better ones.

You just had a conversation with a Large Language Model (LLM) — the technology behind Perplexity, ChatGPT, Claude, Gemini, and dozens of other tools.

Where does generative AI fit?

"AI" is an enormous field. Most of what the military has used AI for — image recognition, predictive maintenance, logistics optimization — is not what we're talking about here. Generative AI is a specific type that creates new content: text, images, code, video, music. It's what happens when you ask a question and get an answer that reads like a human wrote it.

The short video below walks through what generative AI actually is, why it's so capable, and the two ways it can burn you. It runs about five minutes. Watch it before you go on.

How this was made: AI tools drafted the script, Google's NotebookLM turned it into this narrated video, and a person reviewed the whole thing for accuracy before it went in. AI did the work; a human stayed responsible at every step. That is the whole idea behind this read-ahead.

How fast is this changing?

If you tried AI last year and were unimpressed, that's understandable — but the operative words are "last year." AI capabilities are improving at a rate that has no precedent in the history of technology.

You've used AI and you understand roughly how it works. Now let's make it work better. A few simple moves make the biggest difference for someone starting out. They take seconds to learn and immediately improve results.

1. Be specific about what you want

The #1 finding across all the research on AI prompting is brutally simple: vague input produces vague output. This should feel familiar — it's the same principle behind a good operations order. "Go secure that building" produces different results than "Take second squad, clear rooms 1-4 on the ground floor, report when complete."

Notice the specific prompt tells the AI the format (500 words), the scope (amphibious lift, last 3 years), the details expected (ship classes, numbers), and even asks it to flag uncertainty. Every constraint you add makes the output more useful.

2. Show it an example of what "good" looks like

This is one of the most reliable techniques there is, and the most underused. Instead of only describing what you want, hand the AI one good example and tell it to match. Want a storyboard in a particular format? Paste one you have already built. Want summaries in the style your boss likes? Show it one she approved. Across the research, a single concrete example shapes output more than a paragraph of instructions. It is the difference between describing a five-paragraph order and handing someone a well-written one to model.

3. Tell it who the answer is for

This is the honest, useful core of the old "give it a role" advice. Tell the AI who the answer is for — "explain this for a brigade commander," "write this for a brand-new lieutenant" — and it adapts the form of the response: the tone, the length, the level of detail. What it will not do is make the answer more accurate. Naming a fancy expert role adds no knowledge the model was missing, and on hard factual or reasoning tasks it can even backfire. So state your audience to shape how the answer lands, not to make it more correct.

4. Iterate — don't accept the first answer

This is the behavioral shift that matters more than any technique. Most people treat AI like a search engine: one query, one answer, done. The biggest improvement comes from treating it like a back-brief.

You give guidance. You get a response. You correct misunderstandings. You refine. You continue. That's exactly what good iterative prompting looks like. Some of the best AI results come on the third or fourth exchange, not the first. And you are not limited to words: show the AI a screenshot of what it got wrong, tell it what to fix, and it will read the image and correct itself. Most people never realize these tools can see.

If your score went up — and for most people, it goes up significantly — you just proved something to yourself: the quality of AI output is substantially determined by the quality of human input. This is the central insight of everything that follows.

Everything you've done so far is prompt engineering — crafting better questions. But there's a level beyond that: context engineering — giving the AI better information to work with.

What this looks like in practice

Upload a document and ask about it. Paste in a 30-page report and say "What are the three most important things a brigade commander needs to know from this?"

Provide your own background. "I'm a logistics officer planning for a division-level deployment to the Indo-Pacific. I need to brief my commanding general in 48 hours."

Set persistent instructions. Tools like Claude Projects, ChatGPT custom instructions, and Gemini Gems let you set context that applies to every conversation.

Now that you've used AI, watched how it works, and experimented with making it better, here are the terms you'll encounter going forward. These should now have experiential anchors — you've seen most of these in action.

Everything so far has focused on text. But generative AI also builds presentations and turns documents into finished products. The next two modules are optional, but one of these tools might save you hours this week. Start with the things you make most: slides and briefings.

Presentations — This One Will Change Your Monday

If you spend hours building slide decks — and you do, because you're a military officer — this will get your attention.

Interactive briefings, not just slides

The most capable AI tools can now write the code for a full interactive, click-through briefing: the kind you navigate like a web page, with buttons, charts, and even 3D models you can rotate. This read-ahead is one of them. Claude is particularly strong at this, and the other frontier tools are moving the same way. One honest caveat: which AI tools are approved on government networks changes over time, so check what's available to you now rather than counting on any single one.

Turn a Document Into Almost Anything

You have already seen it work: the narrated video back in Module 1 was built in NotebookLM, from a script that other AI tools drafted, with a person reviewing every step.

Generative AI doesn't stop at words and slides. It also makes images, video, and music, and at a pace worth seeing for yourself.

Images — Create, and (More Useful) Edit

Everyone knows AI can make an image from a description. Fewer people know it can edit one, and that is the part that saves you time. Hand it a photo and tell it what to change: remove an object, swap the background, clean up a shot.

A real example. Someone trapped a stray cat and wanted a clear picture to ask around whether anyone recognized it, but the wire trap was in the way. One sentence to Gemini, "remove the cage," produced this:

Same cat, no cage, in a few seconds, with no photo-editing skills required.

For making images from scratch, the field is close and moves fast. ChatGPT is excellent at images that contain text and at conversational, back-and-forth editing. Google's Gemini (its image model is nicknamed "Nano Banana") tops several 2026 quality rankings and has a generous free tier through the Gmail account you may already have. Try one: ask for "a military operations center with officers reviewing a digital map display, photorealistic," then refine it: "make it nighttime, add more screens, more dramatic lighting." That's iteration, applied to images.

Video & Music

Over the last two hours, you've learned how to get better answers from AI. You've seen your own ratings improve. You may have uploaded a document and gotten a useful summary in seconds, or watched Gamma build a slide deck faster than you could open PowerPoint.

But better-sounding answers create a new problem.

Every domain that has adopted automation has discovered the same pattern: when humans rely on automated systems without maintaining their own independent capabilities, performance improves on average but fails catastrophically at the margins. Aviation calls it automation bias. Medicine has seen it. Nuclear energy has seen it. The military will not be exempt.

These aren't prompt engineering problems. They're not even context engineering problems. They're questions about trust, judgment, independence, and organizational design. They're the questions AI 201 is built to help you answer.

Carry that question with you. It's where AI 201 begins.