What makes a coach agentic
A basic roleplay bot just talks. A real human trainer does much more than talk. They pull up a slide, draw a diagram on a whiteboard, walk through a deck, and show you the thing they are describing. The difference between a chatbot and a coach is the ability to take action, to use tools in the middle of a conversation the way a human trainer would.
An agentic coach can do the same, because it has tools. At Tough Tongue AI, the coach can present Google Slides, generate images to set a scene, draw on a canvas to explain a concept, and use a browser to bring in real context. That turns abstract practice into something concrete.
Consider a rep learning to handle a pricing conversation. A basic bot might say “the buyer pushes back on price, what do you do?” An agentic coach pulls up the actual pricing slide from the team’s deck, walks the rep through the three tiers, explains the anchoring strategy behind the layout, and then runs the conversation with the slide visible. The rep practices handling “that is too expensive” while looking at the same visual they will have on a real call.
Or consider competitive positioning. Instead of describing a competitor’s product in text, the coach opens a browser, pulls up the competitor’s pricing page, and walks the rep through the differences in real time. The rep sees what the buyer sees. They practice responding to “I saw on their website that they offer X” with the actual page in front of them.
This is what separates an agent from a script. A script follows a fixed path. An agent observes, decides, and acts. It chooses when to pull up a slide, when to draw a framework on the canvas, when to generate an image of an org chart to explain a multi-threaded deal. The coach adapts its tools to the rep’s needs in the moment, the same way a good human trainer reaches for a whiteboard when a concept is not landing verbally.
Here is a walkthrough of the Google Slides integration, which is how the coach pulls up and presents a real deck mid-conversation:
You can read the full setup details in the Google Slides integration guide.
Why memory changes everything
Memory is the other half of a complete coach. Without memory, every session starts from zero. The coach does not know who the rep is, what they practiced last week, or what they struggle with. It is like seeing a personal trainer who forgets your name and your injury history every visit.
When the agent remembers who a rep is and what they struggled with last time, coaching stops resetting every session and starts compounding. The coach can open with “last time you rushed the discovery and skipped the budget question, let us focus there today.” That is exactly what a good human coach would do.
Memory enables several specific behaviors that make coaching dramatically more effective.
Progressive difficulty. The coach tracks the rep’s performance across sessions and adjusts the difficulty of scenarios accordingly. A rep who has mastered the basic cold call opener gets promoted to a more hostile buyer persona, or a multi-stakeholder scenario. A rep who is still struggling with objection handling gets a simpler scenario with more coaching pauses built in.
Pattern recognition. Over ten sessions, the coach can identify patterns that no single session reveals. “You tend to drop your close when the buyer mentions a competitor” or “your discovery is strong in the first three minutes but you stop asking questions after the buyer gives you good news.” These cross-session patterns are invisible without memory.
Personalized pacing. Some reps learn fast and want to move quickly through material. Others need more repetition. Memory lets the coach match the pace to the individual without a manager having to configure anything. The coach simply remembers how many times a rep has practiced a concept and whether they passed the last scorecard.
Continuity between modes. When the coach remembers what it taught in a coaching session, it can reference that material during a roleplay. “Remember the three-part framework we covered on Tuesday? Try using it now.” This creates a thread between learning and practice that disappears without memory.
Tools make the practice rich. Memory makes it continuous. Together, they create something closer to a dedicated human trainer than to a chatbot with a script.
The three-part mix: roleplay, coach, reverse roleplay
The pattern we recommend building toward is a mix of three modes, not a single roleplay.
1. Roleplay. This is the test and the practice. The rep runs the conversation against an AI buyer who pushes back, goes silent, raises objections, and behaves like a real prospect. The rep gets scored on specific criteria: did they open with a relevant hook, did they ask discovery questions before pitching, did they handle the objection without discounting immediately, did they close with a clear next step.
A roleplay session typically runs 5 to 10 minutes. The rep speaks, the buyer responds, and the conversation unfolds naturally. When it ends, the rep gets a scorecard with specific, actionable feedback. Not “good job” but “you asked two discovery questions, the target is four, and you jumped to the demo before confirming the buyer’s pain.” This is where skill becomes reflex. Repetition with feedback is how you move a technique from conscious effort to automatic execution.
2. Coach. This teaches the concepts before and around the practice. The coach uses slides, browser context, and a canvas to explain frameworks and walk through material. This is where knowledge gets built.
A coaching session might walk a rep through the MEDDPICC framework, using slides to explain each letter, drawing on the canvas to map a sample deal, and then quizzing the rep on how they would apply each element to their current pipeline. Or it might cover competitive positioning by pulling up a competitor’s website and walking through the talk track for each differentiator. The coach is not testing the rep. It is teaching them, building the conceptual foundation that the roleplay will later turn into skill.
3. Reverse roleplay. Here the agent plays the rep and demonstrates the ideal pitch or the best way to handle an objection. The rep watches and listens. They see and hear what good looks like before trying it themselves.
This is the most underused mode, and one of the most powerful. Many reps have never heard a great version of the pitch they are supposed to deliver. They have read the script, maybe watched a recording, but they have not experienced a live demonstration tailored to the specific scenario they are about to practice. In a reverse roleplay, the agent delivers the cold call opener the way a top performer would, handles the “we already have a solution” objection with the ideal response, and closes with a textbook next-step ask. The rep listens, then immediately switches to roleplay mode and tries it themselves. The gap between what they heard and what they delivered is the learning.
Why one mode is not enough
These three reinforce each other. The coach builds the knowledge, the reverse roleplay shows the model of excellence, and the roleplay turns it into a reflex.
A program built on only one of them is incomplete. Only roleplay, and reps practice without ever being taught or shown the standard. They develop habits through trial and error, but those habits may not reflect the best approach. Only coaching, and they learn concepts they never make automatic. They can explain the framework on a whiteboard but freeze when a real buyer pushes back. Only reverse roleplay, and they know what good looks like but have never done it under pressure.
The mix matters for the same reason that athletes watch film, get coached on technique, and then scrimmage. Each mode fills a gap that the others leave open. A typical weekly sequence might look like this: Monday, a 15-minute coaching session on the new competitive talk track. Wednesday, a reverse roleplay where the agent demonstrates the talk track against a skeptical buyer. Friday, a roleplay where the rep runs the scenario themselves and gets scored. By Friday, the rep has learned the concept, seen it executed well, and practiced it with feedback. The talk track is no longer something they read in a doc. It is something they can do.
Building toward the complete coach
The complete coach combines tools, memory, and all three modes into a single system that adapts to each rep over time. Building it does not have to happen all at once. Most teams start with roleplay because it is the most intuitive mode: give the rep a scenario and let them practice. That alone is valuable.
The next step is adding coaching sessions that teach the material the roleplay tests. If your reps are practicing a new pricing conversation, build a coaching agent that walks them through the pricing deck and the anchoring strategy before they roleplay it. The coaching session gives them the knowledge. The roleplay gives them the reps.
Then add reverse roleplay for the scenarios where reps consistently struggle. If your scorecard data shows that 60% of reps fail the competitive objection scenario, build a reverse roleplay that demonstrates the ideal response. Let reps hear it, then practice it. The reverse roleplay closes the gap between “I do not know what to say” and “I have heard what good sounds like.”
Memory ties it all together. As the coach accumulates data on each rep’s performance, preferences, and progress, it becomes less like a tool and more like a trainer. It knows what this specific rep needs to work on next, not because a manager configured it, but because it has been watching.
You can try all three modes right here. Each one runs in your browser, no sign-in required.
Try the roleplay mode. This is a technical cold-call simulation against a skeptical buyer.
Try the coach mode. This coach teaches concepts using slides, canvas, and conversation.
Try the reverse roleplay. Here the AI pitches you, demonstrating the ideal approach so you can hear what good sounds like.
You can explore a full collection of sales coaching scenarios, including roleplays, coaches, and reverse roleplays, in the Sales Coaching Collection.
The same embedding approach drops any of these agents into a lesson, an LMS module, or a training page. A single page can walk a rep through a coaching module, demonstrate the ideal call via reverse roleplay, and then drop them into a live roleplay, all on the same screen.