Google Gemini Spark: The Complete AI Agent Course

Google Spark is in beta, and its features and interface will keep changing. This lesson covers how AI tools update, why new features often reach US and Ultra users first, and how this course is structured. First you learn the basics, then you practice with Drive, Gmail, Keep and Calendar, and finally you work through real case studies for professions like content creator, bookkeeper and teacher.

Configure the Spark settings that matter most. Turn on Personal Intelligence memory so Gemini remembers your context across tasks, and add global instructions that control how Gemini and Spark respond. Then connect your Google Workspace apps and see how non-Google MCP connectors fit in. The lesson also covers running tasks remotely, Keep Awake, and the backup options (backup failure alerts and Precise Backups) that protect your files.

Learn the difference between a regular Gemini chat and a Spark agentic task. You'll build your first multi-step task, which searches the web, adds events to Google Calendar, writes a report and emails it. You'll also learn to read Spark's thought process to find errors, and see why Spark's built-in Google access is faster than external connectors.

Give Spark safe access to your computer through the Gemini desktop app. Learn how linked folders work as a security boundary, why you should connect separate folders for important files, and how to summarize a whole folder's contents. You'll also follow task progress in the sidebar, export responses to Docs or Gmail in one click, and manage, rename, pin and unlink tasks.

Spark works with whole directories, not just single files. Use simple prompts to rename files, create new folders, split a PDF into one file per chapter, and move content between linked folders. You'll also see why Spark can only act inside connected folders, and how to set up access without giving it your entire computer.

Let Spark read and understand the files in a messy folder and propose an organization plan before it changes anything. The plan covers categories, sorting logic and naming conventions. You can review and adjust it, then confirm it, and see how to restore your files from Gemini's automatic backup if something goes wrong.

Recreate the CLAUDE.md-style instructions file in Google Spark. Learn how a local instructions file keeps your workflow consistent across folders, and how it differs from global instructions. You'll also see how to ask Spark to create the file by analyzing your folder and asking you questions, and how to keep improving it with new rules as you work.

Combine Google Workspace and local files in a single Spark task. You'll convert a local PDF to a Google Doc and upload it to Drive, download Google Slides as a local .pptx, and create a Google Task reminder. You'll also learn quick access through the plus menu and how to check what each Google connector can and cannot do.

Go beyond Google services with third-party connectors. Call a connector with the @ symbol, ask what it can do, and add a custom cloud MCP connector using a link or a client ID and secret. You'll see how local and cloud MCPs differ, run a read-only Trivago hotel search, and learn how to judge which connectors are safe to trust.

Learn what Spark skills are and why they're one of Spark's most powerful features. AI models already know how to do the job, but they don't know your process, standards or tone. A skill is the onboarding file that teaches them. This lesson covers why skills only work in Spark and not in regular chat, who can benefit from them, and the two main skill types: document and research skills, and automation and orchestration skills.

Turn any task into a reusable skill with a single prompt, so you never paste the same instructions again. You'll learn how Spark uses a skill's description to run it automatically from a normal conversational prompt. To run a skill on purpose, type "/" to pick it from your list.

Look inside a skill and see its essential parts: name, description and instructions. You'll edit a skill manually and with Gemini, and see the YAML format when you download it. You'll also learn why the name should use kebab case, why the description should be 200 to 600 characters, and how SKILL.md points to bundled assets, scripts and reference files that load only when needed.

Find ready-made skills on SkillHub, SkillsMP and Smithery, and filter them by category or keyword. Download a skill as a ZIP and install it in Spark using the Upload button. You'll install a simple web-research skill and a more advanced social skill that bundles reference and eval files.

Build your own skill with a five-part prompt template: name, purpose, trigger phrases, step-by-step process, and format and tone. You'll create an "ideas-to-lyrics" skill that turns any paragraph into rhyming, singable song lyrics, then test it on a real example. The template is in the section downloads.

Spark normally creates single-file skills, but this lesson shows a workaround for multi-file skills with an assets folder. Using an extended template, you'll build a coffee shop FAQ assistant that answers customer questions using only the FAQ document. You'll package it as a ZIP, upload it manually, test it, and learn how to have Spark draft a skill-creation prompt for you.

Create a skill that uses other skills. You'll build a daily AI news brief that checks TechCrunch, The Verge and Ars Technica, goes deeper on each story with the web-research skill, and compiles a PDF. It then passes that PDF to the social skill to draft posts, and uploads it to Google Drive after you approve.

Learn what subagents are and how they differ from skills. A subagent runs part of a task in a separate parallel instance and returns only the result, which keeps the main task's context window small and efficient. You'll delegate web research to a subagent with a simple prompt, and learn how to ask for subagents inside your own skills.

Have Spark analyze your sent Gmail messages and create a writing-style profile, then use it to draft replies that sound like you. You'll build a reusable email-drafting workflow with instruction files in a working folder, which works much like a project, and then turn the same workflow into a skill. Spark never sends anything without your approval.

Create a "pimp-my-inbox" skill that sorts and labels the last 7 days of email into clear categories, reusing your existing Gmail labels. You'll schedule it to run weekly, see why approval steps can stop scheduled tasks, and build a read-only inbox briefing instead. The briefing lists top priorities, action items, deadlines and emails waiting for a reply.

Three practical Gmail workflows. First, find promotional senders, count opened vs. unopened emails, and label the unread ones for quick cleanup. Second, filter your newsletters down to the ones worth reading. Third, catch up after a vacation by sorting your inbox into urgent, needs a reply, FYI and ignore.

Combine Gmail with other Google services. Pull deadlines, meetings and payment dates from your emails and add the ones you approve to Google Calendar. Then build a contact list from your inbox in Google Sheets with a local Excel copy, update it later without duplicates, and turn the process into an "update-my-contacts" skill.

Create events in a dedicated Google Calendar, then have Spark prepare you for the day. For each meeting it gathers context from the event, related Gmail threads and linked Drive documents, and creates a prep doc with goals, an agenda, key questions and a "Heads up" section that flags conflicting information.

Turn disorganized Google Keep notes into an action plan. Spark reads your quick notes, merges duplicates, skips finished items and groups your to-dos, then creates organized Google Tasks lists with due dates. You'll also rewrite messy notes into clean versions, using a workaround because Spark can't edit existing Keep notes yet.

Clean up a messy Downloads folder automatically. Spark sorts files into category folders, sends uncertain ones to "Review me," and writes an index file. Then you'll process a folder of invoices and credit card statements: extract every expense, organize the files by month and category, and build monthly expense reports in Google Sheets and Excel.

Use Google Sheets like an app. Plan and build a habit tracker with checkboxes and formulas that calculate streaks, completion rates and weekly goals as you tick boxes. Then ask Spark to analyze your progress and find patterns, and see how to apply the same idea to your own job.

Create a monthly budget planner in Google Sheets with formulas that compare planned and actual spending and flag overspending. Paste your rough expense notes and Spark categorizes them for you. Spark can also read invoices from a local folder and split multi-category receipts into separate transactions.

See why AI without context is a guessing machine. Spark writes a quarterly business report from a Google Sheet twice: once with numbers only, and once with company goals, finance rules and meeting notes added. The second report catches missed goals, broken rules and pending decisions, and cites the source of every point.

Turn a report into a Google Slides presentation. First plan the outline, then generate the deck with speaker notes, and refine it with Gemini inside Slides. You'll also see how Google Slides differs from NotebookLM slides, and use a brand style guide to improve an existing deck's content, fonts and colors while Spark flags any contradictions.

Put everything you've learned into practice with real-world case studies. In each one you take on a profession and use Spark to solve the everyday problems that job brings. Every case study comes with an assignment sheet and practice files, so you can try each task on your own first and then solve it step by step with me. The difficulty increases as you go, and every workflow can be adapted to your own work.

Step into the role of a middle school teacher preparing a lesson on the water cycle. Spark turns rough notes into a complete lesson plan using the school's template, then creates a printable quiz with an answer key and an easier version for students who need extra support. You'll also try converting the quiz into an interactive HTML page and learn why you should always double-check generated code.

Grade student work fairly and consistently with AI. You'll build a "grade-with-rubric" skill that includes the school's grading rubric, so Spark scores each answer by criteria, assigns a grade band and writes feedback for the student. When you grade the whole class, it also summarizes common mistakes and suggests what to teach again.

Run the weekly operations of a café. You'll create a "weekly-order" skill that calculates how much to order from past usage, adds a safety buffer, picks the cheapest supplier per unit and drafts the order emails. Then you'll build a drink cost spreadsheet with formulas that checks every product against a 30% cost rule, accounts for an upcoming milk price increase, and suggests new prices.

Spark reads a messy staff chat and builds next week's schedule in a spreadsheet, following the café's rules and flagging conflicts instead of inventing solutions. Then you'll create a "reply-to-reviews" skill with brand-safe rules. It also cross-checks complaints against your stock and staffing files to find the real cause and suggest fixes.

Bring all the café work together. Save chat results as local files, then have Spark combine the supplier order, drink costs, staff schedule and review replies into a one-page Monday briefing covering money, problems and the decisions you need to make this week.

Work as a freelance bookkeeper for a small business client. You'll build a "categorize-transactions" skill that sorts a bank statement into the client's own categories, separates personal purchases, refunds and owner's draws, and flags duplicates and unclear lines. Then you'll match receipts against the bank statement to catch missing receipts, wrong amounts and cash purchases with no bank charge.

Split a single PDF of receipts into individual files, each named by date, vendor and amount, and save them in category folders. Then close the month: total the expenses by category, review the table before anything is written, and fill in the client's profit and loss sheet without breaking its formulas. You'll finish with a plain-language summary for the client that compares the month with the previous one and lists open items.

Act as a YouTube creator on planning day. Spark analyzes channel stats for the last 20 videos, comparing long videos with Shorts and looking past views at % viewed and new subscribers to spot clickbait. You'll then build a "repurpose-video" skill that turns one transcript into 3 Shorts scripts, an Instagram caption and a community post, keeping the creator's style and applying the price correction from the video.

Group viewer comments by what people are asking for, combine them with your performance analysis to choose the next video and three title options, and flag spam and scam comments. Then check a sponsorship offer against your rate card, exclusivity terms, usage rights and channel rules, and get a clear verdict plus what to ask for instead.

Create a one-page media kit to send to brands. Spark uses your channel stats to present who you are, your typical views and % viewed for long videos and Shorts, your best topics, your rates and how you work with brands. It uses honest numbers, so one viral outlier doesn't inflate your typical views.