AI Tutors, Coaches, and Practice Bots: When They Help and When They Don't

Welcome & Purpose Of The Show

Hello, and welcome to the Designing with Love podcast. I am your host, Jackie Pellegrin, where my goal is to bring you information, tips, and tricks as an instructional designer.

AI Scales Practice But Not Design

To make decisions, receive feedback, and try again. AI can help close that gap. But here's the anchor for this episode. AI can scale at practice, but it cannot replace thoughtful practice design. A bot is not automatically a coach. A chatbot is not automatically a tutor. A conversation is not automatically learning. The design still matters. So that's the big shift. AI can make practice easier to create at scale. Now let's ground ourselves in what still has to be true for practice work. What doesn't change is this. Learners still need practice that is meaningful, focused, and connected to real performance. Good practice is not just go chat with the bot. Good practice needs a clear goal, a realistic look, useful feedback, a safe environment, and a path back to real-world application. This matters because AI tools can create the appearance of practice without actually helping learners improve. A learner might have a conversation with a bot and think, that was interesting. But as instructional designers, we have to ask: did they practice the right skill? Did they receive accurate feedback? Did the activity feel confidence? Did it reinforce the right behavior? Did it prepare them for the real situation? That's where the designer stays essential. AI can help create the interaction, but you design the learning conditions around it. And when those learning conditions are missing, AI tutors and practice bots can create some very predictable problems. The risk is not that AI tutors, coaches, and practice bots exist. The risk is using them in situations where they are not well designed, not well supervised, or not the right

Four Risks Of Practice Bots

fit. Here are four common risk areas. Risk number one, the feedback is inaccurate. If a bot gives feedback that sounds helpful but is wrong, learners may leave more confident and less correct. That's a dangerous combination. This is especially risky in areas such as compliance, safety, legal, medical, HR, or policy-heavy training. Risk number two, the interaction becomes too generic. Sometimes the bot gives learners vague encouragement like, great job, try again, or consider your audience. That might feel supportive, but it may not help the learner improve. Good feedback should point to the performance standard and help the learner understand what to adjust. Risk number three, learners overtrust the tool. When a tool sounds confident and responsive, learners may assume it is correct. That's when expectations matter. Learners should know what the bot can help with, what it cannot help with, and when they should check with a real person. Risk number four, the use case is too sensitive. Some situations are not a good fit for open-ended AI practice without careful controls. For example, trauma-informed conversations, disciplinary conversations, mental health support, legal or medical advice, high-stakes evaluation, or private learner or employee data. In those cases, the tool may still have a role, but the guardrails need to be much stronger. So if those are the risks, let's look at the upgrade.

When AI Is A Good Fit

How to decide whether an AI tutor, coach, or practice bot is actually a good fit. Here's the upgrade. Instead of asking, can AI do this? Ask what kind of practice are we designing and what level of risk comes with it? Let's break it down. Best fit number one, low stakes concept support. AI tutors can be useful when learners need help understanding a concept in a different way. For example, explaining a term, comparing two ideas, summarizing a process, generating study questions, or offering additional examples. This works best when the content is low risk and learners are encouraged to verify important information. Best fit number two, rehearsal and role play. Practice spots can be helpful when learners need to rehearse conversations. For example, customer service responses, sales discovery questions, coaching conversations, interview practice, or feedback conversations. This works best when there is a clear scenario, a defined role, and a feedback rubric. Best fit number three, reflection and planning. AI coaches can support reflection when learners need to think through next steps. For example, what did I learn? What would I do differently next time? What is my action plan? Or what obstacle might get in the way? This works best when the bot is not pretending to be a therapist, manager, or expert decision maker. Now, if you're designing any of these, you need five design requirements.

Five Requirements For Safe Practice

Requirement number one, a clear task. What exactly is the learner practicing? Requirement number two, a realistic scenario. What situation should the bot simulate? Requirement number three, a feedback rubric. What does good performance look like? Requirement number four, clear boundaries. What should the bot not do? And finally, requirement pause, take two. And finally, requirement number five, human escalation. When should the learner pause and go to a person? The last one is important. Human escalation is not a failure of the tool, it is part of responsible design. Now let's turn those ideas into a simple decision tool you can use before building or recommending an AI practice experience.

The Help Fit Decision Test

Here's your next move. Use the help fit test. Before you use or design an AI tutor, coach, or practice bot, ask yourself four questions. H is for helpful. What is the tool helping the learner practice or understand? If you cannot name the learning purpose clearly, the tool may become a novelty instead of a support. E is for evidence-based. What feedback, rubric, or standard will guide the response? The bot should not be freestyling feedback. It should be connected to a clear performance standard. L is for low risk or limited risk. What could go wrong if the tool gives poor feedback? If the answer is serious harm, legal risk, safety risk, or emotional harm, you need stronger controls or a different solution. P is for protected. What data, privacy, and permission guardrails are in place. This connects back to our earlier episodes on policies, permission, and data literacy. If learners are entering personal, sensitive, or workplace-specific information, you need to know what is allowed and what is protected. So the Help Fit test is helpful, evidence-based, low-risk, and protected. If you can answer those four questions, you are in a much better position to decide whether AI-driven practice belongs in your learning experience.

Customer Service Role Play Example

Imagine a team wants to use an AI role play bot for customer service training. At first, the request sounds simple. Can we create a bot so learners can practice difficult customer conversations? That could be a great fit, but the learning architect needs to ask a few questions. What type of customer situation are they practicing? What does a good response look like? What feedback should the bot give? What should the bot avoid? When should the learner talk to a supervisor or trainer? Now the design gets stronger. Instead of an open-ended bot that says whatever it wants, the team builds a focused role play. One scenario type, one skill focus, one feedback rubric, one reflection question, and one escalation rule. That is a practice pause to that is a practice bot that helps. But if the bot is used for sensitive employee relations conversations, gives vague feedback, and has no human escalation path, that is not good design. Same technology, different architecture.

Checkpoint Challenge & Companion Resource

Alright, let's make this practical with this week's checkpoint challenge. Here's your checkpoint challenge for the week. Think of one place in your work where learners need more practice. Then run it through the help fit test. Helpful, what skill or decision are they practicing? Evidence-based. What rubric or standard guides feedback? Low risk. What could go wrong? Protect it. What data or privacy guardrails matter? If you cannot answer all four yet, that does not mean don't use AI. It means you have design work to do first. And to make that easier, I created a quick companion resource for this episode. Before you go, I made an interactive companion called AI Practice Compass. It's a quick click-through guide you can use to decide whether an AI tutor, coach, or practice bot is a good fit for your learning experience. If this episode helped you, please follow or subscribe and share it with a designer who is curious about AI-driven practice but wants to use it responsibly. AI tutors, coaches, and practice bots can be powerful when they are designed with purpose. But the goal is not to replace human teaching, coaching, or feedback. The goal is to create better opportunities for learners to practice, reflect, and improve. So before you add a bot, ask, what is the practice? What is the feedback? What are the boundaries?

Aristotle Quote & Closing Thanks

Before I conclude this episode, here's an inspiring quote by Aristotle. For the things we have to learn before we can do them, we learn by doing them. Thanks for spending time with me today. Until next time, keep it practical, keep it human, and keep designing with love. Thank you for taking some time to listen to this podcast episode today. Your support means the world to me. If you'd like to help keep the podcast going, you can share it with a friend or colleague, leave a heartfelt review, or offer a monetary contribution. Every act of support, big or small, makes a difference, and I'm truly thankful for you.