# Gemini > Canvas for turning data into interactive dashboards, Deep Research for multi-step investigations, and the newer modes that change what Gemini is good for _20 min · intermediate · track: chat-agent · id: gemini_ > **Team:** > > Gemini's two standout features are Canvas, where raw data turns into a > working tool in minutes, and Deep Research, where a vague question turns > into a cited brief while you go do something else. You already know the > chat box; this course is those two, plus a few of the newer modes worth > knowing. You already know the model picker, attachments, and connected apps from the [Chat agents](/course/chat-agents) course. The features below are where Gemini stops being a chatbot and starts being a tool that does the work. > **Note:** > > The rest of the bootcamp builds from Claude, not Gemini. If Gemini is not part of your stack, you can skip this course and jump straight to a [Level 3 track](/#level-3) from the skill tree. Most of these features sit behind a Google AI Pro or Workspace Business plan. If you only have free Gemini, this is read-now, try-when-you-upgrade. ## Canvas: turn data into a tool **Canvas is the side-by-side mode where Gemini builds the actual tool while you chat with it.** Attach a file, ask for a dashboard, and watch a working web app appear next to your conversation. You iterate by talking to the result ("add a date filter", "swap the bar chart for a line chart", "color it by region instead"). The point is not the chart itself. The point is that things that used to need a developer ticket now happen in two minutes, by you, on whatever data you happen to have. A messy CSV becomes a usable view. A doc becomes a deck (Canvas exports to Google Slides). A rough idea becomes a clickable prototype you can show somebody. The mental shift: stop thinking of Gemini as a chat that writes code, start thinking of Canvas as the place you build small tools you keep using. The official help page covers how to switch it on and what it can build: [Canvas in Gemini Apps Help](https://support.google.com/gemini/answer/16047321). Building in Canvas is cheap; maintaining a small tool is the same work it has always been. Build a lot, keep only what you use a second time, throw the first version away without regret. > **Tip:** > > **Try it.** Switch on Canvas, attach any messy spreadsheet from your work, and ask: "Show me what I should care about in this data, with three charts that explain it." Iterate by talking to the dashboard until it looks like something you would actually use. ## Deep Research: the long-running investigation **Deep Research is Gemini's mode for the kind of question that would normally cost you forty minutes of tab juggling.** Competitive landscape, regulatory overview, summarising what changed in a field this quarter. You ask the question, Gemini drafts a research plan you can edit (rename topics, drop sources, add angles), and once you approve it, it spends ten to twenty minutes searching the web, reading pages, cross-checking, and producing a structured report with citations. It is not the right tool for "what is X". It is the right tool for "give me a structured brief on X covering Y, Z, and W, with sources I can audit". Use it the way you would use a junior analyst: clear scope, named deliverables, sources you trust. > **Heaven:** You scope the brief tightly, edit the plan before approving, then spot-check two citations when the report lands. Deep Research saved you forty minutes and you can defend every claim. > > **Hell:** You ask 'tell me about AI in HR' and forward the polished report unread. The tool did the work, you put your name on it, and you cannot answer the first follow-up question. > **Tip:** > > **Try it.** Pick a question you would otherwise lose forty minutes researching this week. Switch on Deep Research, scope it tight ("with three vendor examples", "covering Q1 2026 only"), edit the plan before approving, and spot-check two citations when the report lands. ## What else ships with Gemini A few features ride along with Gemini that are not strictly Canvas or Deep Research, but worth knowing exist. - **Audio overviews.** Turn a long report (or any long doc) into a two-host podcast-style summary. Useful for walks, gym, commute. Same feature behind [NotebookLM](/course/notebooklm), now built into Gemini. - **Image and video generation.** Imagen for images, Veo for short video clips. Live in the chat now, no separate tool. Real for "I need a placeholder visual right now", not for production assets. - **Live mode.** Real-time voice or video with the model. Useful for screen-share style "what am I looking at" or "talk me through this code". Niche, but stunning the first time. The [Talking to your AI](/course/talking-to-your-ai) course goes deeper on when live voice is and is not the right tool. > **Tip:** > > **Try it.** Pick the one above you have never used. Spend five minutes finding it in the Gemini interface and running one real prompt against it. The point is not to use it tomorrow, the point is to know it exists when the right job comes along. ## Quiz **Q1.** You are the CFO. You have a 12-month CSV of tool spend and you want to spot the weird months yourself before the end of the day. You only have Gemini. What is the strongest play? - a. Run Deep Research with a prompt to investigate tool-spend anomalies and benchmarks; let it produce a structured brief. - b. Paste the CSV into chat and ask Gemini to describe the patterns and flag the months that look off. - c. Attach the CSV in Canvas and prompt for an interactive dashboard with monthly trend, year-over-year comparison, and a filter by category. Iterate live until you see what you need. **(correct)** - d. Ask Gemini to generate a one-page summary of common tool-spend anomaly patterns you can apply mentally to the data. _Explanation:_ You do not want a brief, you want to look at the numbers and react. That is exactly the seam Canvas fills: data in, interactive view out, iterate by talking to it. Deep Research is for web investigation, not your own private spreadsheet, and would burn twenty minutes producing the wrong artefact. Plain chat gives you a static description you cannot poke at. A generic summary of anomaly patterns is the inverse of what was asked, it does not touch the data. ## Hands-on 1. Start a new chat. Click **Add files** to attach a CSV (any spreadsheet from your work; if you have nothing handy, ask Gemini to "generate 20 rows of sample sales data: date, product, region, revenue, units_sold" and save the result). Pick **Canvas** from the **Tools** menu and prompt: ``` Build an interactive dashboard from this data: charts for trends over time, a breakdown by region, and a date filter. Add a summary card at the top with total revenue. ``` When the dashboard appears on the right, iterate twice ("change to dark mode", "swap to a line chart"). Notice how fast it is. 2. Open a new chat, switch on **Deep Research** from the **Tools** menu, and ask one question that would normally cost you a half hour of tabs. Examples: ``` Brief me on how AI agents are being used in HR tech in 2026, with three concrete vendor examples. ``` Or: ``` Summarize regulatory changes affecting EU SaaS in the last 6 months. ``` Read the proposed plan first. Edit it before approving. Skim the final report when it lands. Click two of the citations to confirm Gemini is not making things up. ## Reflect - Where in your week are you doing slow research that Deep Research would handle while you do something else? Pick the next one and put it on the calendar. - What is one small internal tool (a dashboard, a tracker, a quick interactive view of some data you keep in a spreadsheet) you have always wanted but never asked a developer to build? Could Canvas get you a usable v1 today?