Chat, Artifact, Cowork, Codex: A Plain-English Map of What These AI Things Actually Are
You are not behind.
The words are confusing, and most people explaining them skip the part where they tell you what the words mean.
Here’s what that looks like in practice. You sit down for a training that promised to be beginner-friendly. Three minutes in, someone says “just drop that into an artifact and connect it to your MCP.” You nod. You write it down. You keep watching, but you’re no longer really following, because the whole thing is now built on a word you never got a definition for.
Then the demo starts, and every new thing gets stacked on the thing you already missed.
That isn’t a you problem. That’s a teaching problem.
So this post has one job: define every term before it gets used. No demos. No workflows. No assuming you already know. Just the words, in plain English, one at a time. And yes I did use AI to help me write it (I gave it the goal of the post, expanded the outline and then edited and fact checked the draft AI came up with and added in my own examples).
Why This Is Genuinely Confusing
A few things happened at once.
These tools launched fast, often in beta, often to some plans before others. Features got renamed. The same word means different things at different companies. Something available on your friend’s account may not have rolled out to yours yet.
And they live in different places. Some things only work in a desktop app. Some only in a browser. Some work everywhere but behave a little differently in each.
People who use these tools every day still mix the terms up. I do it. Nothing is wrong with you.
Let’s go through them.
A Chat
A chat is the back-and-forth text box. You type something, the AI types back, you keep going.
This is what most people mean when they say “I use ChatGPT” or “I use Claude.” You ask a question, you get an answer, you maybe ask a follow-up.
Think of it as: a conversation. It’s very good at answering, explaining, brainstorming, and rewriting. It ends when you close the window.
Everything else in this list is a variation on “what if it could do more than talk.”
An Artifact
An artifact is a thing the AI makes that lives in its own window, next to the conversation instead of buried inside it.
When Claude writes you something substantial, a document, a checklist, a small tool, a chart, it opens in a panel to the right of your chat rather than scrolling past in the conversation. You can edit it, ask for changes, flip between versions, copy it, or download it.
Claude decides to make an artifact when the content is self-contained, more than about fifteen lines, and something you’d plausibly want to reuse or come back to.
Think of it as: the difference between talking about a recipe and being handed the recipe card.
Now, the question everybody actually asks: does it save automatically?
Sort of, and this trips people up, so here’s the precise answer.
The artifact stays in that conversation. Go back to the chat, and it’s still there.
But it does not automatically show up in your Artifacts library in the sidebar. To put it there, you open the artifact and click “Publish.” That’s the step people miss, and it’s why some folks think artifacts disappeared on them.
Live Artifacts vs Static Artifacts
Claude now has two kinds of artifacts, and this is worth separating because they behave very differently.
A regular artifact is the thing Claude makes inside a chat: a document, checklist, chart, bit of code, small app, or reusable piece of content that opens next to the conversation. It can be edited, downloaded, copied, published, or shared depending on your plan. It lives with the chat unless you publish it or access it through the Artifacts area. It is also static and doesn’t automatically update.
For example, I have a static artifact that I provide access to that allows users generate a customer support skill for their business. Static artifacts can be compared to Custom GPTs, as they can be similar.
A live artifact is different. Live artifacts are created in Claude Cowork, only in the Claude desktop app, and they are built to keep working over time. Think dashboards, trackers, reference pages, comparison tools, or status views that can refresh with current information from your connected apps or local files. Keep in mind that anything connected to outside tools should be treated like giving the AI access to that specific bridge
For example, I have a bingo card live artifact with my goals that I can check off and it logs when certain boxes got checked etc.
So the plain-English distinction is this:
- A regular artifact is a static reusable thing Claude made.
- A live artifact is a reusable thing Claude made that can keep updating.
For example, if Claude writes you a checklist, that is a regular artifact. If Claude builds you a dashboard that checks your calendar, Airtable, or project list and refreshes when you open it, that is a live artifact.
Final Note on Artifacts
One more thing to check: artifacts require “Code execution and file creation” to be switched on in Settings, then Capabilities. If you’ve never seen an artifact, that setting is the first place to look.
And the sharing question: if you build (and publish) an artifact that uses AI inside it and share it with other people, their usage counts against their Claude subscription, not yours. Sharing it costs you nothing, whether ten people use it or ten thousand.
Claude Chat, Claude Cowork, and Claude Code
These are three different ways of working with Claude. They are not three different companies or three different subscriptions. Here’s the honest difference.
Claude Chat is the conversation, as described above.
Claude Cowork is a desktop app, though it has started to roll out as a browser feature as well. CoWork is the same Claude, but able to actually do things. Instead of answering one message at a time, it takes on a whole multi-step task, works through it, and hands you finished output. Documents, spreadsheets, organized files, research pulled together.
Chat and Cowork now share one home. You’ll find both in the same message box: pick “Cowork” to hand over a task, pick “Chat” to have a conversation. You don’t go somewhere else.
Claude Code is the same underlying capability built for programmers, run through a terminal. While this may seem like something you wouldn’t ever use, if you are interested in vibe coding or building apps using tools like Lovable, starting in Claude Code or Codex (which we’ll get to more later) can be a great place to build it out before you use up all your Lovable credits.
It tends to be a little more generous in usage than the actual vibe coding tool. If there is ever something you need to execute in Terminal (which is something on a Mac), then it can walk you through step by step: click here, copy this, paste that. It is still something you can use. I wouldn’t say I use it as often, but I do absolutely use it.
Where Cowork runs, since this is the most common question
- It’s available on paid plans only: Pro, Max, Team, and Enterprise. There’s no free Cowork.
- It works in the Claude desktop app on macOS and Windows.
- It also works on the web at claude.ai and in the mobile apps, though those are in beta and rolling out gradually, starting with Max plans. On the Pro plan I only have it as a desktop app.
- Your session runs remotely on Anthropic’s servers, which means work keeps going when you aren’t at your laptop.
- But: if the task needs a file on your actual computer, or your browser, you need the desktop app open and your computer to be “awake” so Claude can reach it.
ChatGPT, Work, and Codex
Same basic split, different company.
ChatGPT is the conversation.
Codex is OpenAI’s agent that goes and does the work. Worth knowing plainly: Codex is positioned as being built for code. But I use it for the same things as I do Claude CoWork. And plenty of other people will use it for file wrangling and automation that isn’t strictly programming.
That being said, Chat GPT recently rolled out Chat GPT Work.
I’d compare Chat GPT Work to Claude CoWork, and then Codex aligns with Claude Code.
But unlike CoWork, Chat GPT Work is available both in the desktop app and via browser, via a toggle:
Here’s a difference that matters for your wallet: Codex is included across ChatGPT plans, including the free tier. Usage limits vary by plan, but there’s a free on-ramp. Claude’s Cowork has no free tier at all.
Desktop vs. Browser
Short version: the browser version runs on the company’s servers. The desktop app can also reach things on your own computer.
If you want the AI to open a folder on your laptop, read your files, or click around in your browser for you, you need the desktop app installed and open. If you’re fine uploading files manually and working in the cloud, the browser is enough.
Both are the same account and the same subscription. It’s not two products.
Custom GPTs
A Custom GPT is a version of ChatGPT that somebody set up in advance with specific instructions, uploaded reference files, and chosen tools. You can build your own or use one from the GPT Store.
Think of it as: a saved starting point. Instead of explaining what you want every time, the explaining is baked in.
Two things worth knowing. First, most people using “someone else’s custom GPT” are using a general-purpose helper that knows a topic, not something that knows their business. Second, ChatGPT’s memory feature does not work inside custom GPTs that you don’t own.
Projects
Both platforms have something called Projects, and they mean roughly the same thing: a container for related work.
In ChatGPT, a project groups chats, files, and context under one objective.
In Claude, projects do the same, and inside Cowork and Codex a project is a workspace with its own files, instructions, and memory.
Think of it as: a folder that remembers things. Instead of every conversation starting from nothing, work inside a project shares context.
A limitation worth knowing: what Claude remembers about you in regular chat does not currently carry over into Cowork. Inside Cowork, memory works within projects.
Scheduled Tasks
A scheduled task is work that runs on a timetable without you starting it.
Both platforms have this. In Claude Cowork you type /schedule inside a task, or use the Scheduled section in the sidebar. ChatGPT has scheduled tasks too, one-time or recurring.
Think of it as: a standing instruction. “Every Monday morning, check this and write me a summary.”
Here are some examples of scheduled tasks I have running:
- External Event Inbox Monitor: discover summits, bundles, giveaways, contributor events, virtual events, and similar collaboration opportunities mentioned in Gmail, then enrich the COH Speaker Alumni Airtable base with event and speaker evidence for future pitching.
- pin board assignment: Weekly: assign best-fit Pinterest boards to pins missing them — all brands (COH, LDMW, SCH, ES)
- Footer copywriter: Daily: write Broadcast Footer copy (Notes field) in Liz’s voice for records with complete info and empty Notes — using the record’s link, swipe copy, and promo dates. Never touches Kit; the existing API push handles that.
- Customer Support AI Draft Review – Hub: review any open Customer Support tickets and draft a response for Liz to review and finalize
This is usually the feature that makes people sit up, because it’s the first one that isn’t you asking for something.
Skills, Plugins, and Connectors
These three get mixed up constantly. Here’s the distinction.
A connector links the AI to another tool you already use, like your email, your calendar, or your database. It’s a bridge to an outside service you’re using. Think:
- Google Drive
- Clickup
- Airtable
- Promo Hub HQ
A skill teaches the AI a repeatable way of doing something. If there’s a specific process you want followed the same way every time, a skill captures that. For example, I have a skill that drafts my JV workshop Registration pages in Wordpress. The only thing that needs to be swapped out on the template is the partner I’m working with and AI can handle that.
A plugin is a bundle. In Claude Cowork, a plugin packages skills, connectors, and sub-agents together into one installable thing. On the OpenAI/ChatGPT side, a plugin can include skills, apps, and app templates.
Think of it as: a connector is a phone line, a skill is a set of instructions, and a plugin is a starter kit containing both.
In Claude Desktop these live together under a “Customize” section, which is a genuinely useful thing to know when you’re looking for them.
And since this comes up: skills are built on an open standard called Agent Skills, which means they’re designed to work across AI platforms rather than being locked to Claude or ChatGPT alone.
A Note on What Connectors Do Not Do
What you can do with a connector often depends on the connector itself and you can usually decide what kind of permissions you want to allow.
Connecting your email to an AI does not hand over your account.
A connector gives specific, limited access to specific things. It doesn’t log in as you, take over, or start acting on its own. Nothing gets sent without you. You can disconnect at any time. You also choose how often it stops to ask permission before doing anything, and the default is to ask.
You stay the one who hits send.
Paid Plans and Usage Limits
The thing nobody explains until you hit it.
Claude Pro is $20 a month. It gives you at least five times the usage of the free tier, plus access to Claude Code and Cowork. Your usage resets on a rolling session basis every five hours, and there’s also a weekly cap across all models. You can see your reset time in Settings, then Usage.
Three things that actually help with your usage:
Use plain chat for simple questions and save Cowork for real multi-step work.
Batch related work into one session instead of starting fresh repeatedly.
Check Settings, then Usage, so you know your reset schedule instead of being surprised by it.
Pro subscribers can also turn on usage credits to keep working past the included limit. Max plans give you more room if you need it.
On the OpenAI side, Codex usage draws from a shared pool with several other features, and how much a task consumes depends on its size and complexity. I rarely hit the ChatGPT/Codex usage limits and on the rare occasion I do, I usually have at least one usage reset available to me.
So When Would I Use Which?
| What you’re doing | What you want |
|---|---|
| Asking a question, thinking out loud, getting something explained | Chat. Cheapest, fastest, plenty. |
| Handing over a whole task and stepping away | Cowork (Claude) or Work/Codex (OpenAI). This is where it stops answering and starts doing. |
That’s the whole decision. Most people should be in chat more than they think, and in Cowork less often but for bigger things.
Model Names, Briefly
When you see a dropdown asking you to pick a model, here’s the short version. Please note that this section is accurate as of the time of writing, Aug. 2, 2026 but that the models available update and change regularly.
Claude
These are all Claude; the naming looks intimidating but the choice is simple.
Fable 5 is the frontier model. Reach for it when the thinking itself is the hard part, note it’s only available on the Max plan or above to use it as a Pro user, you’ll have to purchase usage credits.
Opus 5 is close to Fable’s intelligence at half the price. A sensible default for demanding work and importantly available to those on the Pro plan
Sonnet 5 is the agentic workhorse: strong at reasoning, tool use, and getting real tasks done. This is the one most people should sit on most of the time.
Haiku 4.5 is the fast, cheap one. Good for simple, high-volume things where speed matters more than depth.
If you don’t want to think about it: stay on Sonnet, and move up when a task feels genuinely hard.
Open AI/Codex/Chat GPT
These are all all ChatGPT, again the naming may look intimidating but the choice is simple.
In regular ChatGPT, the model picker is now mostly plain-English: Instant, Medium, High, Extra High, and Pro.
Instant is the fast everyday option. It is the default for normal questions, quick drafting, explaining, rewriting, and the kind of thing where you do not need the AI to stop and think for a long time.
Medium and High are the reasoning options. Use these when the task is more complex: planning, strategy, troubleshooting, analysis, or anything with several moving parts.
Extra High is a deeper reasoning option available on higher plans. Reach for it when the answer really needs care.
Pro is the highest-capability option for the hardest tasks and longer workflows. It is slower and uses more of your allowance, so I would not use it for ordinary questions.
Under the hood, the names are more technical. GPT-5.5 Instant powers fast everyday ChatGPT responses, while GPT-5.6 Sol powers the reasoning options on eligible plans. GPT-5.6 Sol Pro powers Pro. You may also see GPT-5.6 Terra and GPT-5.6 Luna in Work, Codex, or the API, but they are not selectable in normal ChatGPT conversations.
If you do not want to think about it: stay on Instant for everyday chat, move to Medium or High when the task needs reasoning, and save Pro for the genuinely hard stuff.
That’s the Vocabulary
Nothing here required you to already know anything. That was on purpose.
If you read this and thought “oh, that’s all those words meant,” that’s the point. The concepts underneath are not complicated. The naming is.
Now, the part this was written for.
On Wednesday, August 12 at 12:00 pm Pacific / 3:00 pm Eastern, I’m running a free live workshop called The System Behind My $40/Month AI Setup. It’s about what happens when Claude and ChatGPT both actually know your business, so you’re not re-explaining yourself every time you open a new chat, and you’re not paying $100 a month for one tool trying to do everything.
Now that the words make sense, come see the whole thing working together.
Here is one final breakdown of all the vocab:
| Claude language | OpenAI / ChatGPT language | What it means in plain English | Availability |
|---|---|---|---|
| Claude Chat | ChatGPT | A normal back-and-forth conversation | Claude: Free and paid. ChatGPT: Free and paid. |
| Artifact | Canvas | A separate workspace for writing, code, or reusable output | Claude Artifacts: Free, Pro, Max, Team, Enterprise. ChatGPT Canvas: Free and paid plans. |
| Claude Cowork | ChatGPT Work | The AI takes on a longer task and produces a deliverable | Claude Cowork: paid Claude plans only, starting at Pro. ChatGPT Work: limited desktop-app access on Free and Go; expanded access on Plus and Pro across desktop, web, and mobile. |
| Claude Code | Codex | Agent-style work with files, code, terminal/browser support | Claude Code: paid Claude plans, starting at Pro. Codex: included across ChatGPT plans, including limited access on Free and Go, with expanded usage on paid plans. |
| Project | Project | A container that keeps chats, files, and instructions together | Claude Projects: Free and paid, with Free limited. ChatGPT Projects: Free and paid, with more capacity/features on paid plans. |
| Skill | Skill | A saved repeatable process the AI can follow | Claude Skills: Free, Pro, Max, Team, Enterprise. ChatGPT Skills beta: listed across Free, Go, Plus, and Pro; workspace/personal skill controls may vary by plan and admin settings. |
| Connector | App / Connector | A bridge to another tool, like Gmail, Drive, Airtable, or ClickUp | Claude: varies by connector and plan. ChatGPT Apps: available across plans, but specific apps and actions can depend on plan, region, and permissions. |
| Plugin | Plugin | A bundle of skills, tools, connectors/apps, or templates | Claude Plugins: included with Cowork, so paid Claude plans only. ChatGPT Plugins: available across Free, Go, Plus, and Pro, with actual use depending on the plugin, app permissions, plan, region, and workspace settings. |
| Scheduled task | Scheduled task | Work the AI runs later or on a recurring schedule | Claude Scheduled Tasks: inside Cowork, paid Claude plans only. ChatGPT Scheduled Tasks: Plus, Pro, Business, and Enterprise. |
| Model picker | Model picker | Where you choose speed vs deeper reasoning | Claude: available model options vary by plan. ChatGPT: basic model access exists on Free and Go, but advanced reasoning models start at Plus, with Pro getting the highest-capability options. |
Sources
Every fact in this post was checked against official documentation as of August 2026. Features change fast, so if something reads differently when you get there, the docs win.
