UFT-CWE Teacher Workshop · Reference Handout
A map of what is actually on your screen, what each part does, and what the thing cannot do.
This is a reference, not a lesson. Read it once through and then keep it open beside the app. The goal is that nothing on your screen is a mystery — that when you see a word in the sidebar, you know what it means, whether you have it, and whether you need it.
Two honest framings before we start. First: most of what follows you will never use. That is fine. The core of this tool is a text box. Everything else is scaffolding around that text box, and you can add scaffolding when you hit a wall, not before. Second: the software changes faster than any handout can track. Where something here has drifted, the official pages linked at the bottom are the authority — not this page, and not Claude's own account of itself, which is often out of date about its own product.
Contents
Open the app and the left-hand column looks roughly like this. If yours is missing items, that is normal — the sidebar shows only what your plan and your device support.
The top five are places you can be. The bottom group is history. The circle at the very bottom is you.
The default. A conversation. This is where the overwhelming majority of your work will happen and there is no shame in never leaving it.
Each chat is its own container. Claude does not know what you said in a different chat unless you turn on memory or ask it to search your history (see §3). Start a new chat when you change subjects; keep one going when you're building something across days.
A folder with a memory and a standing instruction. You upload reference material once — a syllabus, a rubric, a style guide, three sample lessons — and write instructions that apply to every chat inside that project. Then each new chat starts warm instead of cold.
This is the single highest-leverage feature for teachers, and it is the one almost nobody sets up. If you are going to build one thing, build a project for the class you teach most.
What are projects? · Projects are available on every plan, including free; free accounts are capped at five. Custom project instructions are a paid-plan feature. See personalization features.
Two things share this name, which is the source of most confusion.
The panel. When Claude makes something substantial — a worksheet, a chart, a working web page — it opens in a window to the right of the chat instead of scrolling past in the message. You can edit it in place: highlight text, click Edit with Claude, say what you want changed.
The sidebar section. A gallery of artifacts you've kept. Note the trap: artifacts do not appear here automatically. You have to open one and click Publish. Everything else lives only inside its original chat.
Artifacts require Code execution and file creation to be switched on under Settings → Capabilities. If your artifacts aren't working, that toggle is the first thing to check.
Claude Code — a separate workspace for software development. Each session gets its own folder, its own history, its own copy of a codebase. It is a genuinely different product wearing the same coat.
For our purposes: you can ignore this tab. It is there because the same app serves programmers. Nothing you need for a GED packet lives behind it. (I use it; you do not have to.)
The newest and strangest item. Dispatch is a long-running agent: you describe an outcome once, it breaks the work into tasks, runs them in the background, and tells you when they're done. You don't watch it work. You come back to the result.
It runs on your own computer, which means your machine has to be awake and the desktop app open. Paired with the phone app, it turns your phone into a remote control — text your desk a task from the bus, find the work waiting when you sit down.
Two honest notes. It requires a Pro or Max plan and the desktop app on macOS or Windows. And on genuinely complex tasks it is unreliable — roughly coin-flip, by most accounts. It is very good at fetching, summarizing, and organizing; it is not yet good at judgment.
Dispatch and its parent, Cowork, give an AI agent permission to read, move, and delete real files on your real computer, and to act inside your connected accounts. That power is the point, and it is also the risk: a bad instruction, or a manipulated one picked up from a web page along the way, produces real consequences that can be hard to undo. Give it a dedicated folder. Do not give it the run of your machine.
History. Starred is manual — you pin what you want to find again. Recents shows the last handful. All chats › opens the full archive, which is searchable.
A discipline worth adopting early: rename your chats. The auto-generated titles are serviceable for a week and useless after a month.
Settings, your plan, and your account live here. See §3.
The chat screen is mostly one text box. Around it:
Give it the real thing. The gap between "write me a GED reading passage about work" and "here is the passage I used last term, here is the rubric, here are two student responses — write me a third at the same level" is the entire difference between a toy and a colleague. Context is the whole game.
Click your initials, bottom left. The parts that matter:
Account-wide standing orders. Applied to every conversation, everywhere. "I teach adult ESOL and GED. Assume adult learners. Never use em dashes. Ask me one clarifying question before long tasks." Write this once and every chat improves.
Chat search and memory · Import and export your memory
Memory drifts. After a couple of months it is carrying one-off corrections you made about one file, stale role descriptions, and project facts leaking into unrelated chats. Read the summary occasionally and prune it. An assistant optimizing for a version of you from March is worse than one starting fresh.
On consumer plans (Free, Pro, Max), conversations may be used to improve the models unless you turn that off. The toggle is in Settings → Privacy. Know where it is and make your own decision; do not let it be made by default.
The same account, the same conversations, on different surfaces:
Where can I access Claude? · Beta and research preview features
Not just text in a chat window. With file creation enabled:
Create and edit files with Claude · Visual and interactive content · Can Claude produce images?
This section is not a disclaimer. It is the part of the handout with the highest practical value, and I would rather you read it than §5.
Claude will produce a citation that does not exist, a statistic that was never measured, a court case that never happened — in the same confident register it uses for things that are true. There is no tell. The prose does not get worse when it starts making things up. Anything you would be embarrassed to be wrong about, verify.
Incorrect or misleading responses
It will produce a link that goes nowhere. It will say it emailed something, or saved a file, when no such thing occurred. Click the link. Check the folder.
Training data stops at a fixed point. Ask about anything after it and you get either a search result or a confident guess — and the model is not always careful about telling you which. For anything current, make it search.
How up-to-date is the training data?
Push back on a correct answer and it will often fold. This is the failure mode that should worry teachers most, because it is invisible: you will feel like you won the argument. If you want a real second opinion, ask it to argue the other side before you say what you think.
Left alone it drifts toward a generic register — competent, sanded-down, slightly American, slightly corporate. Every distinctive thing in your teaching that survives contact with this tool survives because you insisted on it. That insisting is the work. It does not go away.
A few things worth settling before you paste anything in.
Student work and student data. Our learners are adults, many of them immigrants, some with immigration status they would not want written down anywhere. A student essay pasted into a chat is a student essay on someone else's server. Strip names. Strip identifying detail. Ask whether the pedagogical gain is worth it, case by case — and understand that the answer is sometimes plainly yes and sometimes plainly no.
Union and grievance material. Same logic, higher stakes. Incognito exists for a reason.
Know your plan's terms. Consumer and organizational plans handle data differently. If UFT-CWE supplies your account, the org's retention and export policies apply — an administrator can export conversations, including incognito ones.
Who can view my conversations?
Anthropic's own courses are free, require no account with them beyond an email, and are better than most of what's on YouTube. They're video-based and self-paced.
Everything in this handout is scaffolding around a conversation. The teachers who get the most out of this tool are not the ones who learned the most features — they are the ones who kept insisting on their own standards until the machine met them. Nothing here substitutes for knowing what a good lesson looks like. It just gets you to the fourth draft faster.
Prints clean black on white — no background fills, easy on the ink and the photocopier.