Hands-On AI Exercises

info

This content was developed with AI assistance and is regularly reviewed for accuracy.

Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.

Reading about AI only gets you so far. The exercises below turn the AI Fluency lessons into skills you can use at work tomorrow, and each one ends with a clear way to check your result.

In short: This is a hands-on practice library for people who already use AI and want better results. It has quick 15-30 minute exercises, mini-projects, and larger projects covering briefing, verification, choosing a mode, connecting your accounts safely, content creation, research, and agents. A free AI account is enough for most of it; the exercises say when a paid plan helps.

Pick an exercise

Coming back to continue? Jump straight to the level that fits the time you have today.

If you have...	Start here	You will practice
15-30 minutes	Exercise 1: Prompt Engineering Fundamentals	Writing and refining requests
15-30 minutes	Exercise 2: AI Tool Comparison	Choosing the right tool
15-30 minutes	Exercise 3: Quality Control Practice	Fact-checking AI output
20-30 minutes	Exercise 4: Delegate a Task to an Agent	Supervising multi-step AI work
20-30 minutes	Exercise 5: Write a Brief	The six-field brief
20-30 minutes	Exercise 6: Verify Against a Planted Error	Catching mistakes on purpose
20-30 minutes	Exercise 7: Choose a Mode	Fast, thinking, research, or agent
20-30 minutes	Exercise 8: Connect One Connector	Narrow permissions and review
1-3 hours	Mini Projects	Building a personal AI workflow
5-10 hours	Major Projects	Research, automation, teaching others

New to the track? Exercises 5 to 8 follow the lesson order: Brief, Check, Decide, and Supply Context. Exercises 1 to 4 work with no reading at all.

Quick practice exercises (15-30 minutes each)

Exercise 1: Prompt Engineering Fundamentals

Goal: Learn the basics of asking an assistant for what you want

Task: Write requests for these scenarios and test them with your chosen AI tool:

1. Content Creation

   • Write a professional email declining a meeting request
   • Create a social media post for a local bakery's new product
   • Generate a product description for an eco-friendly water bottle

2. Analysis and Research

   • Summarize a news article in 3 bullet points
   • Compare pros and cons of electric vs. gas cars
   • Explain a complex topic (like cryptocurrency) to a 10-year-old

3. Creative Tasks

   • Write a haiku about your favorite season
   • Create a short story opening with a specific character and setting
   • Brainstorm 10 unique business names for a pet grooming service

For each one, try a bare version first ("write a bakery post"), then a version that names the audience, the purpose, and the format. Notice which details changed the output most.

Success Criteria:

• Each request produces relevant, useful output
• You can refine requests to improve results
• You understand how specificity affects quality

When these feel easy, move to Exercise 5, which replaces one-line prompts with a full brief.

Exercise 2: AI Tool Comparison

Goal: Understand when to use different AI tools

Task: Complete the same task with 3 different AI tools and compare results:

The Task: "Create a week-long meal plan for a vegetarian family of 4 with a $150 budget"

Tools to Test:

• ChatGPT (free or paid)
• Claude (via claude.ai)
• Google Gemini (at gemini.google.com)
• Bonus: Any specialized food/nutrition AI tool

Compare:

1. Quality and practicality of meal suggestions
2. Budget accuracy and calculations
3. Consideration of nutritional balance
4. Presentation and organization of information
5. Additional helpful details (shopping lists, prep tips, etc.)

Note which tier you used for each. Free tiers run the smaller model of each family, so a weak answer may say more about the tier than the brand; the Decide lesson explains why.

Deliverable: Create a simple comparison chart noting strengths and weaknesses

Exercise 3: Quality Control Practice

Goal: Develop skills for verifying and improving AI output

Task: Generate content with AI, then apply quality control

1. Generate: Ask AI to write a 300-word article about the benefits of remote work
2. Review: Check for:
   • Factual accuracy
   • Logical flow and organization
   • Tone consistency
   • Grammar and spelling
   • Completeness of information
3. Improve: Rewrite or ask the AI to fix any issues found
4. Verify: Cross-check facts with reliable sources

Success Criteria:

• You can identify potential problems in AI output
• You can improve content through editing or a better request
• Final content meets professional standards

Exercise 4: Delegate a Task to an Agent

Goal: Practice handing a multi-step task to AI and supervising the result

Exercises 1-3 are conversations: you ask, the AI answers, you ask again. An AI agent works differently. You describe an outcome, and the agent plans the steps, searches, reads, and assembles the result on its own. Your job shifts from writing prompts to giving a clear brief and checking the work.

Task: Use any assistant that offers an agent, research, or "deep research" mode (these are usually on paid tiers). Give it this brief, filling in the brackets:

"Research [3 tools for a task you do every week, such as scheduling social posts]. For each one, find the price of the cheapest paid plan, whether there is a free tier, and one limitation real users mention. Put the results in a table and list every source you used."

While it runs, notice what the agent does on its own: the searches it chooses, the pages it opens, and any point where it asks you a question.

When it finishes, spot-check the output:

1. Open two of the cited sources and confirm the prices match
2. Look for any claim with no source attached
3. Ask a follow-up that changes the scope, such as "Add a fourth tool that works offline"

Success Criteria:

• Your brief described the outcome, not the individual steps
• You verified at least two facts against the original sources
• You can name one thing the agent did well and one thing you had to correct

Ready to go further? Build Your First Agent walks through agent tasks step by step.

Exercise 5: Write a Brief

Goal: Turn a vague request into a brief an assistant can act on without guessing

Task: Pick a real task you will do this week, such as a status update, a client email, or a one-page plan.

1. Before: Type the request the way you normally would, in one line. Save the result.
2. Interview: In a new conversation, write "Before you begin, ask me the questions you would need answered to do this well," followed by your one-liner. Answer its questions.
3. Brief: Using what the interview surfaced, fill in the six fields and attach any source files:

Objective: [what done looks like, and why it matters]
Context: [who I am, who this is for, the situation]
Resources: [attached files / project / connector / URLs; which one is authoritative]
Constraints: [length, tone, out of scope, do not change, approval needed before...]
Output: [format; one example if format matters]
Verification: [how to check; what to cite; what to show me]

4. After: Run the brief in a fresh conversation and compare the two results side by side.

Success Criteria:

• Your brief names a resource and says which source is authoritative
• Your verification field asks the assistant to show evidence, not just assert
• The "after" result needed fewer corrections than the "before" result
• You saved the brief somewhere you can reuse it

Learn more in Brief: How to Ask AI for What You Want.

Exercise 6: Verify Against a Planted Error

Goal: Practice the verification routine on output you know contains a mistake

Checking is easier to learn when you know there is something to find. In this exercise the assistant hides an error and you hunt for it.

Task:

1. Plant: In a new conversation, send this, filling in a topic you know reasonably well:

Write a 250-word explainer on [topic] for a general reader. Include four specific figures or dates and one citation to a real source. Insert exactly one subtle factual error. Do not tell me where it is until I type "reveal".

2. Hunt: Apply the five-minute routine from Check. Trace every figure, open the citation and read the relevant passage, and mark each claim as confirmed, wrong, or unverifiable.
3. Second opinion: Open a fresh conversation, paste the explainer, and ask: "Review this for factual errors and unsupported claims. List each one with your reasoning." Note what it catches and what it misses.
4. Reveal: Return to the first conversation and type "reveal".

Success Criteria:

• You found the planted error, or can say exactly which check would have caught it
• You opened the citation instead of judging it by its title
• You compared your findings with the fresh reviewer's and noted any differences
• You counted any additional errors the assistant made without being asked

That last box matters. If you found a mistake beyond the planted one, you have just seen why every number and citation gets traced.

Exercise 7: Choose a Mode

Goal: Match a task to fast, thinking, research, or agent mode, and see the difference for yourself

Task:

1. Predict: For each task below, write down which mode you would choose and why, using the table in Decide:
   • Reword a paragraph to sound friendlier
   • Work out which of three phone plans is cheapest for your usage over two years
   • Compare five project-management tools for a 10-person nonprofit, with sources
   • Turn a folder of meeting notes into a summary document and a follow-up email draft
2. Test: Take the third task and run it twice: once in the default fast mode, once in research or deep-research mode (or with effort or thinking set to high if research is not on your plan).
3. Compare: Time taken, number of sources, whether the sources are current, and how many claims you could verify.

Success Criteria:

• You can explain your mode choice for each of the four tasks in one sentence
• You recorded the time and source count for both runs
• You can name one task where the slower mode was worth it and one where it was not

Exercise 8: Connect One Connector

Goal: Give an assistant live access to one account with the narrowest useful permission, use it, and review what you granted

Task: Choose one personal account, such as your calendar. Do not connect a work account unless your employer has approved the tool.

1. Find the setting: Look for connectors in your assistant's settings. The name varies by app (Connectors in Claude, Apps in ChatGPT, Apps or Extensions in Gemini), and connectors are available on Claude's free tier.
2. Read the permission screen: Before approving, write down what it asks for. Can it only read, or can it also create, edit, send, or delete?
3. Use it: Ask a question you can check: "What does my Thursday look like, and where could I fit a 45-minute prep block?"
4. Verify: Open the calendar itself and confirm the answer, as Check recommends for anything an assistant read or did on your behalf.
5. Audit: Ask the assistant to list the tools this connector gives it. Then check your account provider's security page (for example, your Google Account's third-party access list) to see the grant from the other side.
6. Decide: Keep it, narrow it, or disconnect it.

Success Criteria:

• You can state exactly what the connector can read and what it can change
• The answer matched your real calendar, or you found and noted the discrepancy
• You found the grant in your account's security settings and know how to revoke it

Before connecting anything that can send or delete, read Protect. For a deeper setup guide, see Connect Your First MCP Server.

Mini projects (1-3 hours each)

Project 1: Personal AI Assistant Setup

Goal: Create a customized AI workflow for your daily needs

Tasks:

1. Identify Use Cases: List 5 regular tasks where AI could help you
2. Tool Selection: Choose the best AI tool and mode for each task
3. Standing Context: Create a project with short standing instructions, and save a brief for each use case
4. Workflow Design: Plan how AI fits into your daily routine, including which connectors (if any) each task needs
5. Testing: Use your AI assistant for one week and document results

Example Use Cases:

• Morning news summary
• Email composition assistance
• Meeting agenda creation
• Learning new topics
• Meal planning

Deliverable: A personal AI toolkit with briefs and workflows you'll actually use. The playbook Set Up Your Personal Agent takes this further.

Project 2: Content Creation Pipeline

Goal: Build an end-to-end content creation process using AI

Scenario: You're creating content for a small business blog

Tasks:

1. Topic Research: Use AI to generate 10 blog post ideas
2. Content Planning: Choose 3 topics and create detailed outlines
3. Writing: Use AI to draft one complete blog post from a brief
4. Editing: Apply quality control and human editing
5. Optimization: Create search-friendly titles and meta descriptions
6. Social Media: Generate social media posts to promote the content

Tools You'll Use:

• AI for ideation and writing
• Manual research for fact-checking
• Your creativity for personalization

Deliverable: One polished blog post plus promotional materials. See Make Things: Writing and Documents for the workflow.

Project 3: Problem-Solving with AI

Goal: Use AI to tackle a real challenge in your life or work

Choose Your Challenge:

• Organizing a complex project or event
• Learning a new skill efficiently
• Improving a process or workflow
• Making a difficult decision
• Solving a creative problem

Approach:

1. Define: Clearly describe your challenge and desired outcome
2. Research: Use AI to gather information and perspectives
3. Brainstorm: Generate multiple solution approaches
4. Analyze: Evaluate pros/cons and feasibility of options
5. Plan: Create a detailed action plan
6. Execute: Implement with AI assistance where helpful

Documentation: Keep notes on how AI helped and where human judgment was essential

Major projects (5-10 hours each)

These are open-ended. For a structured version whose deliverable is a reusable project, skill, or scheduled agent, with a rubric, see Projects: Build Something With an Agent.

Project A: AI-Powered Research Report

Goal: Conduct comprehensive research on a topic using AI assistance

Choose a Topic (examples):

• The impact of AI on your industry
• Sustainable living practices for urban dwellers
• Investment strategies for beginners
• Current health and wellness trends

Process:

1. Research Planning (1 hour)

   • Define research questions
   • Identify information types needed
   • Plan which steps go to a research mode and which you do yourself

2. Information Gathering (3-4 hours)

   • Run a research-mode pass on each question
   • Generate comprehensive outlines
   • Identify reputable sources to verify
   • Create detailed notes and summaries

3. Analysis and Synthesis (2-3 hours)

   • Use AI to identify patterns and insights
   • Compare different perspectives
   • Draw evidence-based conclusions

4. Report Creation (2-3 hours)

   • Structure findings clearly
   • Write with AI assistance
   • Include verified facts and sources
   • Create executive summary

Deliverable: 2,000-3,000 word research report that demonstrates thorough investigation. The playbook Use AI as a Research Assistant covers citation checking in detail.

Project B: AI Workflow Automation

Goal: Automate a repetitive process using AI tools

Example Scenarios:

• Social media content planning and creation
• Customer service response templates
• Data analysis and reporting
• Event planning and coordination
• Learning and skill development tracking

Steps:

1. Process Mapping (1 hour)

   • Document current manual process
   • Identify automation opportunities
   • Decide whether it needs a fixed workflow or an agent
   • Define success metrics

2. Tool Selection and Setup (2-3 hours)

   • Research available AI tools
   • Test different options
   • Create accounts and learn interfaces

3. Workflow Design (2-3 hours)

   • Design new automated process
   • Create briefs and templates
   • Plan integration points and approval steps

4. Implementation and Testing (2-3 hours)

   • Build the automated workflow
   • Test with real scenarios
   • Refine and optimize

5. Documentation (1 hour)

   • Create process documentation
   • Calculate time/cost savings
   • Plan maintenance and updates

Deliverable: Working automated workflow plus documentation. Start with Workflows vs Agents, then follow the worked example in Tutorial: Build an Inquiry Router With an AI Classifier.

Project C: AI Educational Content Creation

Goal: Create educational content that teaches others about AI

Content Options:

• Tutorial video or presentation
• Beginner's guide article or blog series
• Lunch-and-learn session for your workplace
• Social media educational campaign
• Podcast episode or interview

Requirements:

1. Audience Research (1 hour)

   • Define your target audience
   • Understand their current knowledge level
   • Identify their needs and concerns

2. Content Planning (2 hours)

   • Choose specific AI topics to cover
   • Create detailed content outline
   • Plan interactive elements or examples

3. Content Creation (4-5 hours)

   • Use AI to help research and draft content
   • Add personal insights and experiences
   • Create engaging examples and exercises
   • Design visuals or supporting materials

4. Review and Refinement (2 hours)

   • Test content with target audience members
   • Gather feedback and make improvements
   • Ensure accuracy and clarity

5. Publication and Sharing (1 hour)

   • Publish or present content
   • Gather audience feedback
   • Plan follow-up or additional content

Deliverable: Published educational content that helps others learn about AI. If you are teaching a team, Build an AI Literacy Program has a structure to borrow.

Assessment and reflection

Skills Self-Assessment

After completing exercises and projects, rate your progress:

Basic AI Usage (1-5 scale):

• I can write a brief that an assistant acts on without guessing
• I know when to use fast, thinking, research, or agent mode
• I can verify and improve AI-generated content
• I follow privacy and ethical guidelines for AI use
• I can troubleshoot common AI problems, including when to restart

Advanced Skills (1-5 scale):

• I can design multi-step AI workflows
• I can connect accounts with narrow permissions and review what an agent did
• I can help others use AI tools well
• I can evaluate new AI tools quickly
• I can integrate AI into complex projects

Professional Application (1-5 scale):

• I regularly use AI in my work/studies
• I can recommend AI solutions to others
• I understand business implications of AI
• I can measure ROI of AI implementations
• I stay current with AI developments

Learning Reflection

Document Your Journey:

1. Biggest Insights: What surprised you most about AI capabilities?
2. Practical Impact: How has AI changed your daily work or activities?
3. Challenges Overcome: What difficulties did you face and how did you solve them?
4. Future Applications: Where do you see the most potential for AI in your life?
5. Teaching Others: How would you explain AI benefits to someone new?

Next Steps Planning

Based on your experience, choose your path forward:

Continue Learning:

• Build an agent of your own in the AI Agents track
• Focus on applications in your field with the AI and Your Career guides
• Follow developments in What's New in AI

Apply Your Skills:

• Implement AI solutions in your workplace
• Share knowledge with colleagues or friends
• Start an AI-related side project
• Help your team adopt AI with AI for Teams and Leaders

Build Expertise:

• Specialize in specific AI tools or domains
• Seek AI-related responsibilities at work
• Compare formal courses and credentials in the AI Certifications Directory
• Build a portfolio of AI projects

Additional practice resources

Daily Practice Ideas

Week 1: Foundation Building

• Day 1-2: Try 3 different AI tools for the same task
• Day 3-4: Rewrite two of your regular requests as six-field briefs
• Day 5-6: Focus on quality control and fact-checking
• Day 7: Reflect and plan week 2

Week 2: Application Building

• Day 1-2: Create an AI workflow for a work task, with a project holding its standing context
• Day 3-4: Use AI for a creative project
• Day 5-6: Help someone else learn to use AI
• Day 7: Document your progress and insights

Share and Learn

Share Your Learning:

• Post about your AI experiences, including what went wrong
• Show a colleague one brief that saved you time
• Ask questions in the forums and groups you already use for your profession

Learn from Others:

• Follow AI practitioners in your field
• Read case studies and practical examples
• Watch tutorial videos and demos, then try the technique yourself the same day
• Browse the courses listed in Learning & Education

Continued Skill Building

Monthly Challenges:

• Try a new AI tool each month
• Complete one significant AI-assisted project
• Teach one person about AI usage
• Turn your best brief into a reusable project or skill

Quarterly Reviews:

• Assess your AI skill progression
• Update your AI toolkit, projects, and connector permissions
• Set new learning goals
• Plan advanced learning paths

Key takeaways

• Practice on real tasks; every exercise here accepts your own work in place of the sample
• A brief, a verification habit, a deliberate mode choice, and a narrowly scoped connector are the four skills that compound fastest
• Planting an error and hunting for it is the quickest way to make checking a habit
• Read the permission screen before you connect anything, and know how to revoke it

Next steps

Continue to: Projects: Build Something With an Agent

• Revisit the lessons behind these exercises from Start Here: You Already Use AI
• Walk through one complete real task in the playbook Your First Delegated Task