7 AI terms you must know

This week I gave a dataviz talk at a top research institute. I covered dataviz theory, but also the tech side of the job.

(btw, you can hire me to teach dataviz to your team 😋)

Before my "AI for dataviz" demo, I asked the audience about their current AI setup. The vast majority were still just using chatbots.

There is nothing wrong with that. But I believe a few concepts are worth knowing, even if you never become a power user.

So in today's issue, I explain 7 AI concepts in plain words:

1️⃣ Agent

An agent is an AI that actually does things for you. Unlike a chatbot, which only sends text back in response to your prompt.

For instance, Claude Code is an agent that edits your script directly. Nothing to copy-paste. It also reads all the files in your project, and with that context its work is far better than a chat that knows nothing about your code.

If you're curious how agentic coding works for dataviz, check out the live demo I made recently!

2️⃣ Skill

Let's say you never want em-dashes in your writing. Or you want every chart to use #69b3a2, your company's main color. Or there's a task you repeat 3 times a day.

You should write a skill: a small text file with custom instructions for your AI.

For example, I have a skill that teaches my AI how to add a chart to dataviz-inspiration.com. It explains where to store the PNGs, which writing style to use, the possible categories, the mistakes to avoid...

Result: I give my AI some very messy content, and it does everything the way I want, without me repeating myself every time. ☺️

3️⃣ Harness

The AI model is the engine. Just a function that takes text as input and returns text. The harness is the car built around it.

It is the software that gives the model its hands: it can read your files, run code, search the web, and keep working step by step until the task is done. Claude Code, Cursor or Codex are all harnesses.

In my opinion, this is where most of the innovation happens these days. I'll probably talk more about it soon as it's the current hype on X.

4️⃣ Token

AI doesn't read words, it reads tokens: small chunks of text. In English, a token is roughly ¾ of a word.

Everything is counted in tokens: what you pay, how fast the answer comes, and how much text the AI can "remember" in one conversation.

5️⃣ Artifact

A chatbot gives you text. An artifact is a real thing the AI builds for you: a webpage, an interactive chart, a small app, a document.

For dataviz folks, this is a game changer. Ask for "an interactive bar chart of this CSV with a tooltip," and instead of code to paste somewhere, you get a working chart you can open, tweak and share with a link.

6️⃣ MCP / Connector

By default, your AI lives in a bubble. It can't see your emails, your Google Drive or your database.

MCP (Model Context Protocol) is a standard that lets AI plug into your other tools. If an app has a MCP, then your AI can communicate with it, without you pressing buttons.

Most apps call these "connectors."

Once connected, you can ask things like "find last month's survey results in my Drive and chart them," and it just does it.

7️⃣ Vibe coding

It means building software by describing what you want in plain language, and letting the AI write all the code, often without even reading it.

A few months ago, I used to criticize vibe coding a lot. But times are changing, and models keep getting better.

The other day, I was talking with the best developer I know. He doesn't write or read his code anymore. Vibe coding looks like it's becoming the norm, and data visualization won't be spared.

Hope this helps!

If you're still only using AI as a chat: can I ask you why you don't try agents? This is a genuine question, so drop me an email. I read every answer and I'm eager to help.

PS: there is a new dataviz challenge running! It's powered by DataStory and I'm a judge. One more opportunity to work on your viz skills! ❤️

Best,
Yan