Use and modify the prompt at the bottom of each section to go deeper with any AI tool, at your own pace. Change the framing, add your own context, or ask it to go further — the prompt is a starting point, not a script.
Tokens.
The currency of every AI interaction. Tokens are how cost, speed, and limits are measured — understanding them changes how you think about every conversation you have with AI.
When you type a message to an AI, it doesn't read words the way you do. It breaks everything down into tokens — chunks of roughly three to four characters, or about three-quarters of a word. "Understanding" becomes two or three tokens. A full page of text is around 400–500 tokens. The model doesn't see words, sentences, or documents. It sees a stream of tokens.
This matters because tokens are how everything gets measured:
- Cost. Most AI providers charge per token consumed (input) and per token generated (output). A 400-word structured prompt costs more to run than a 10-word casual question. When you're running dozens of tasks a day, this adds up.
- Speed. Generating more tokens takes longer. Long, detailed outputs — especially from advanced reasoning models — can take noticeable time. Shorter, focused requests get faster responses.
- Context limits. Every AI session has a fixed window of how many tokens it can hold at once — your messages, the AI's responses, any files you've attached, and the system instructions behind the tool. When the window fills up, earlier parts of the conversation fall away. This is covered in detail in Context below.
- Model capability. Different models have different token limits and different costs per token. An advanced reasoning model might cost 10–20× more per token than a lightweight model. Matching the model to the task is where experienced users save both time and money.
You don't need to count tokens manually. But understanding that tokens are the currency of AI interactions helps you reason about cost, speed, and limits in a way that "I sent a message" doesn't.
Prompt: go deeper on tokens
Use and modify this in any AI tool:
Explain tokens in AI like I'm a smart professional who has never written code. Cover: what tokens actually are and how they relate to words, why they matter for cost (input vs output pricing), speed, and context limits, how different models have different token costs and capabilities, and include a practical example showing how the same request can use very different amounts of tokens depending on how it's written. Keep it concrete — real numbers, real tradeoffs.
Markdown.
A simple way to add structure to plain text — headings, lists, bold, links — that AI models understand natively. The common language between you and every AI tool.
Markdown is a set of simple text conventions that add structure without complexity. A # makes a heading. A - makes a bullet point. ** makes text bold. That's most of it.
The reason it matters for AI work is that every major AI model reads and writes Markdown fluently. When you structure your prompts with clear headings and lists, the AI produces better, more organised output. When it responds, it uses Markdown to structure what it gives you.
Learning Markdown takes about ten minutes and means you're communicating with AI in its native format. It's also the format behind most modern documentation and note-taking tools like Notion and Obsidian. Once you see it, you see it everywhere.
Prompt: go deeper on Markdown
Use and modify this in any AI tool:
Teach me Markdown in under five minutes. Start with the six things I'll use 90% of the time, show me what each one looks like when typed and when rendered, and then give me a practical exercise: a short document I can write in Markdown right now to practice. Assume I've never used it before but I'm comfortable with technology.
Context.
This might be the single most important concept for becoming an effective AI user. Context is everything the AI can see — and it's finite. Managing it well is where casual users and power users diverge most visibly.
Every AI session has a context window — a fixed amount of space, measured in tokens, that holds everything the model can work with at once: your messages, the AI's responses, any files you've attached, and the system instructions behind the tool. The quality of AI output is almost entirely determined by the quality of context you give it.
When you start a fresh session, the context is empty. As the conversation progresses, every message you send, every response the AI gives, every file you attach — all of it accumulates. Once the context is full, the model starts losing access to earlier parts of the conversation. This is why long sessions go wrong. A model that seems confused, starts contradicting itself, or produces increasingly generic output isn't "getting dumber" — it's working with degraded information.
Practical rules that experienced users follow:
- Start fresh sessions for new tasks. Don't carry a morning's worth of conversation into an afternoon task.
- Watch your context usage. Most serious tools show you how full your context is. Treat 50% as a soft ceiling — beyond that, quality starts to degrade.
- Front-load the important information. What you put at the beginning and end of your context is recalled more reliably than what sits in the middle.
- Every turn is cumulative. A session of 40 exchanges has a very different context profile to a session of 5. This is why experienced users start clean sessions aggressively.
- Large context windows ≠ better. Some models advertise 1 million token windows. Technically impressive, but models often lose coherence in the middle of very long contexts. Focused ≠ limited.
In the workshop, the detailed brief you wrote wasn't just creative writing — it was context engineering. Every detail you provided gave the AI better material to work with. The skill isn't prompting — it's knowing what to put in the room and when to clear the room.
Prompt: go deeper on context
Use and modify this in any AI tool:
Explain context windows in AI as if I'm a professional who uses AI regularly but wants to get significantly better results. Cover: what context actually means technically and how it's measured in tokens, why the same prompt can produce wildly different results in different conversations, what happens when context fills up and why sessions degrade, the 50% rule for context usage, when to start a fresh session versus continuing, and why huge context windows aren't automatically better. Give me five specific techniques for managing context more deliberately, with real examples — not abstract principles.
Skills.
Reusable, shareable packages of knowledge, rules, and assets — written in plain text — that give any AI tool a specific capability without you re-explaining anything. One of the most important structural developments in AI tooling, and most non-engineers have no idea they exist.
In the workshop, you wrote a brief and the AI made decisions about colours, layout, and tone. A skill takes that further: instead of describing what you want each time, you package the decisions once — brand colours, typography rules, tone of voice, logo assets — and the AI applies them automatically on every future request. Think of it as the difference between giving someone verbal instructions every morning and handing them a reference manual they can consult themselves. The AI doesn't forget, doesn't drift, and doesn't need reminding.
A skill is a plain-text Markdown file — typically named SKILL.md — that contains a description of what it does, step-by-step instructions for how to execute the task, references to supporting assets like templates or examples, and metadata that controls when and how it activates. You don't need to write code. Anyone who can write clearly can write a skill — a journalist, a marketing director, a finance analyst, an HR manager. The skill encodes your domain expertise into a reusable workflow.
In late 2025, Anthropic published the Agent Skills specification as an open standard. Within months it was adopted by OpenAI, GitHub Copilot, Google Gemini, Cursor, JetBrains, and dozens of other platforms. Skills aren't one vendor's feature — they're the industry standard for packaging reusable AI capabilities. The brand skill on this site is a working example: the same zip file produces consistent, on-brand outputs across Claude Code, Codex, and Gemini without any of them being told the brand rules in the prompt.
What makes skills powerful at scale is composability. One skill can call another. An agent can select from multiple skills depending on the task. Skills can be combined into workflows. The intelligence isn't in any single prompt — it's in the architecture of skills working together, each drawing on your organisation's domain knowledge. This is how organisations move from individual AI use to systematic AI capability. One person writes the skill. Everyone in the team benefits.
Go deeper: agentskills.io (the open standard) · Agent Skills Masterclass — The AI Daily Brief (Apple Podcasts, Spotify)
Prompt: go deeper on skills
Use and modify this in any AI tool:
Explain AI agent skills as if I'm a business leader who wants to scale AI usage across a team. Cover: what a skill actually is (the SKILL.md file format and the open standard at agentskills.io), how it differs from just writing a good prompt, why non-engineers can and should write them, how skills compose — one skill calling another to build workflows, what kinds of organisational knowledge are worth encoding as skills, and give me three concrete examples of skills that would save a team significant time. Be practical, not theoretical.
This article is a companion to the Vibe Coding for Everyone workshop from Ringier Future Summit 2026. More resources at futuresummit.dylanharbour.com.