Understanding and Working with AI
Understand what AI can do, use it well, and know when to question the result.
12 modules
Self-paced
Certificate of Completion
Enrolled
About this course
AI can be useful, fast, and convincing. That does not make every answer correct. This course teaches you how to work with it without handing over your judgment.
You’ll give it real tasks, check its answers, research questions, work with numbers, write and edit, handle email, compare options, organize work, and build repeatable ways of getting things done.
You’ll also see where it fails, when to trust the result, and when human judgment must make the final call.
Work on real problems
Work with emails, messages, meeting notes, tables, schedules, research excerpts, customer requests, business situations, and project requirements.
Understand what is happening
Understand why answers change, why context matters, why false information appears, and how to verify responses systematically.
What you’ll learn
- Explain in ordinary language how modern AI systems produce answers.
- Decide whether AI is appropriate for a task before using it.
- Give clear instructions, context, limits, and examples.
- Recognize unsupported claims and check important information.
- Research a question without confusing generated text with evidence.
- Draft and edit writing while protecting the underlying facts.
- Write clearer emails and messages for different situations.
- Compare options without handing the final decision to the system.
- Work with tables and numbers while checking the calculations.
- Use AI for everyday workplace tasks without creating more work.
- Understand how AI becomes part of a larger product.
- Build a repeatable process with clear points for review.
- Recognize information that should not be shared.
- Review AI-assisted work before another person sees it.
Curriculum · 12 Modules
Every module includes guided lessons, practical assignments, and an independent challenge.
-
1
Ask the same question twice: Give the same task more than once. Compare the answers to identify what remains stable and what changes.
-
2
See what context changes: Give the same question once by itself and once with useful background information. Compare the results to see how context changes the answer.
-
3
Prediction is not memory: Give several short incomplete passages with possible continuations. Explain in ordinary language how a language model generates text.
-
4
Decide what context matters: Give one task and several pieces of background information. Identify which context is essential and which is distracting or irrelevant.
-
5
Know what it is good at: Give ten different tasks. Classify each one: Use it, Use it and check, Use another tool too, Do not use it. Explain why.
-
6
Find the edge of its knowledge: Use obscure facts, recent information, missing information, and questions the system cannot reasonably answer. Identify where it knows and where it starts guessing.
-
1
Say what you actually need: Start with a vague request like 'Help me with the event.' Turn it into a request with a clear result.
-
2
Give the missing context: The first request does not include enough information. Identify what the system actually needs.
-
3
Set useful limits: Add requirements such as audience, length, deadline, format, things to include, and things to avoid.
-
4
Show what good looks like: Give the learner a useful example and a poor example. Use both to make the expected result clearer.
-
5
Ask for the right shape: Use the same information to produce an email, a checklist, a table, and structured fields. Decide which output fits the task.
-
6
Fix a request that failed: Receive a poor answer. Identify what was wrong with the original request and improve it rather than starting over randomly.
-
1
Find the unsupported claims: Compare the answer with the supplied evidence. Mark anything the evidence does not support.
-
2
Separate fact from inference: Classify statements as supported fact, reasonable inference, opinion, or unsupported claim.
-
3
Check the numbers: Verify percentages, totals, differences, and comparisons. Use deterministic verification for numerical data.
-
4
Check the source: The learner gets several source references. Some support the claim, some do not, and one does not exist. Find the problem.
-
5
Check another way: Verify an important conclusion using the original numbers, another supplied source, or a direct calculation. Do not merely ask the same system, 'Are you sure?'
-
6
Repair the answer: Rewrite the brief so that every important claim can be defended. The evaluator compares factual content with canonical source material.
-
1
Understand the message before replying: Read an email thread with several participants. Identify what happened, what is being asked, what needs clarification, and who needs a reply.
-
2
Write the short version: Receive a bloated AI-written email. Reduce it to what the recipient actually needs while keeping the necessary facts.
-
3
Change the tone without changing the message: Write the same update for a customer, a colleague, and a manager. The facts stay the same; the tone and level of detail change.
-
4
Handle a difficult reply: A customer is unhappy. Draft a response that acknowledges the problem, states what can be done, and avoids unauthorized promises.
-
5
Ask for the missing detail: Receive an incomplete request. Write a brief question asking for the missing piece without annoying the sender.
-
6
Review before sending: Receive an AI-drafted reply that contains subtle errors and unauthorized assumptions. Audit the draft, fix the issues, and approve the final version.
-
1
Start with a real question: Turn a broad topic into a question that can actually be investigated. Give the learner the actual decision context.
-
2
Judge a source: Compare sources with different authors, dates, evidence, interests, and levels of authority. Decide what each source is useful for.
-
3
Gather evidence: Pull the information relevant to the research question. Do not write the conclusion yet.
-
4
Keep the evidence attached: Create research notes where every important claim remains connected to its source.
-
5
Handle disagreement: Two credible sources disagree. Explain the disagreement instead of hiding it.
-
6
Write what the evidence supports: Produce a short research brief. Do not make the conclusion stronger than the evidence allows.
-
1
Draft from real information: Create a first draft using supplied facts. Nothing important should be invented.
-
2
Write for the reader: Turn the same information into versions for residents, volunteers, and local partners. Preserve the facts while adapting for the reader.
-
3
Make writing clearer: Edit a dense piece of writing. Remove repetition, simplify awkward sentences, and improve the order while keeping the meaning.
-
4
Protect the facts: Receive a polished rewrite containing subtle factual changes. Find and repair them.
-
5
Critique before rewriting: Before asking for another version, identify what is actually weak in the existing draft. Then make targeted changes.
-
6
Make the final choices: The system suggests several edits. Decide which ones improve the writing and which ones should be rejected.
-
1
Decide what matters: Define the criteria before comparing the options. Establish meaningful operational criteria from the case.
-
2
Compare the choices: Build a useful comparison using only supplied information. Objective facts should be checked deterministically.
-
3
Find the hidden assumptions: Review an initial recommendation and identify assumptions that were never established.
-
4
Change the priorities: Change the importance of cost, travel, capacity, or accessibility. See whether the recommendation changes.
-
5
Challenge the leading option: Make the strongest fair argument against the current favorite. Engage with real weaknesses, not contrarianism.
-
6
Make the decision: Submit the final recommendation and explain why. AI may support the analysis, but the learner owns the conclusion.
-
1
Understand the table: Identify fields, units, dates, categories, missing values, and limitations before drawing conclusions.
-
2
Ask a useful question: Turn 'What does this data say?' into questions the data can actually answer.
-
3
Check a calculation: Calculate totals, averages, percentages, and changes. Verify important figures independently.
-
4
Find patterns without inventing causes: Identify what changed. Do not claim why it changed unless there is evidence. Distinguish observation from causal speculation.
-
5
Investigate unusual results: Look at outliers. Decide whether they may be errors, unusual days, or genuine patterns using context.
-
6
Write the data brief: Separate what the data shows, what it might suggest, and what cannot be concluded. Deliver an authoritative summary.
-
1
Prepare for the meeting: Review an agenda, participants, and background notes to generate a focused 1-page meeting prep brief.
-
2
Turn discussion into action: Turn messy meeting notes into clear decisions, action items, owners, and deadlines.
-
3
Set the priorities: Work through a multi-task backlog with dependencies. Triage what must happen today versus what can wait.
-
4
Plan the project steps: Draft a sequential project roadmap with buffers and risk checkpoints. Plan around realistic dependencies.
-
5
Prepare the handoff: Create a handoff brief for a teammate taking over a workstream. Ensure zero knowledge gaps.
-
6
Close the loop: At the end of the week, turn updates, decisions, and unfinished work into a manager update, next actions, and open risks.
-
1
The model and the product are different things: Compare a raw model response with a product that adds rules, company information, formatting, and limits. Identify what the product changes.
-
2
Some instructions outrank others: Work through situations where product rules, user requests, and supplied information conflict. Decide which controls the result.
-
3
Context has limits: The system cannot use every piece of information equally. Choose what matters for a particular task from a large set.
-
4
Give it the right information: Compare a general answer with one grounded in an approved company knowledge set. Explain plainly why products supply information.
-
5
Let another system do what AI should not: Study an assistant that can look up an order, calculate a refund, and check an account. Decide what the model handles and what comes from tools.
-
6
Make the result predictable: Define a consistent structure for support requests so another part of the product can use the output.
-
1
Understand the current process: Map what happens now. Do not add AI before understanding the work and its existing bottlenecks.
-
2
Choose where it belongs: For every step, decide whether it should be handled by AI, a simple rule or system, or a person. Justify the assignment.
-
3
Protect private information: Review messages containing sensitive information. Decide what should not be sent to an outside AI service and sanitize it.
-
4
Do not let untrusted text control the process: A customer message includes instructions intended to manipulate the system. Treat the message as DATA rather than trusted instructions.
-
5
Decide where a person must approve: Find high-impact decisions, uncertain cases, exceptions, and irreversible actions that require human review.
-
6
Test before trusting: Run ordinary cases, incomplete cases, unusual cases, misleading cases, and hostile cases through the workflow. Record failures and improve it.
-
1
Understand the assignment: Determine what must be produced, what information exists, what is missing, and where AI could genuinely help.
-
2
Organize the evidence: Separate facts, numbers, opinions, requests, assumptions, and open questions from the dossier.
-
3
Research the open questions: Use the supplied source environment to investigate open questions. Keep claims tied to evidence.
-
4
Analyze the options: Compare several possible plans. Check assumptions and important numbers against constraints.
-
5
Produce the work: Create the recommendation, supporting brief, required communication, and action plan. AI can support; you own the result.
-
6
Audit everything: Check facts, sources, numbers, privacy, claims, tone, requirements, and unanswered questions before submitting.
Certificate of Completion
Earn an official Practisi credential
Upon completing all 12 modules and the Final Challenge with a passing score of at least 80%, you receive an official Certificate of Completion.
- Permanent public credential verification link
- Cryptographically secure verification record
- Directly shareable on LinkedIn and professional portfolios
Learner reviews & ratings
No ratings yet
Frequently asked questions
Do I need technical or engineering background?
No. The course focuses on practical, real-world workflows, clear communication, and output evaluation without requiring programming.
What models or tools are used?
Practisi integrates state-of-the-art foundation models with contextual guidance from Clare to test and verify responses directly.
Can I learn at my own pace?
Yes. All lessons are self-paced, and your progress is saved continuously.
Can I earn a certificate?
Yes. Complete all 12 modules and Final Challenges with a score of at least 80% to earn your Certificate of Completion.