Mark — Valence. Mark walks through how Nadia works under the hood and demos a complete coaching session — from a returning user picking up a previous conversation, to exploring root causes, to building a structured agenda, to role-playing the situation with live line-by-line feedback.

How Nadia Works: AI Agents, Pattern Recognition, and Role-Play

Nadia is not a single AI model — she is a coordinated system of specialized AI agents, each handling a different dimension of the coaching relationship. Separate agents provide empathy, ask reflective questions, track behavioral history and patterns across sessions, and an editor-in-chief layer removes biases and hallucinations before responses are delivered. All of this coordination happens in a fraction of a second, producing something that feels dynamic, conversational, and genuinely coach-like. This walkthrough shows exactly how that works in practice — from a returning user picking up a previous conversation, to role-playing a difficult feedback scenario with line-by-line guidance.

Key Takeaways

• Nadia is built from multiple specialized AI agents working in concert. Separate agents handle empathy, reflective questioning, behavioral memory and pattern tracking, and a bias and hallucination removal layer that acts as an editor-in-chief. All coordination happens in a fraction of a second to produce responses that feel thoughtful, consistent, and coach-like — not like a generic AI chatbot.
• No prompt engineering required — casual, natural language works. Users do not need to structure their input carefully or speak formally. Nadia transcribes accurately and understands natural, conversational language, lowering the barrier to engagement and making sessions feel like real conversations.
• Pattern recognition surfaces insights users haven't yet noticed about themselves. As conversations accumulate, Nadia identifies recurring behavioral themes across sessions — for example, flagging that a user has brought up challenges with giving feedback multiple times. She surfaces this as a developmental observation the user can choose to explore, creating a coaching relationship that becomes more valuable over time.
• Role-play delivers line-by-line feedback in real time. When a user wants to practice a difficult conversation, Nadia takes on the persona of the other person — responding in character — and provides both in-the-moment guidance after each message and a holistic debrief at the end. Users arrive at real conversations having already practiced them.
• Built-in accountability tools keep development moving after the session. Alongside the coaching, Nadia generates action item lists, offers to email a session summary, and can schedule a follow-up debrief. The session becomes the start of a development loop, not a standalone interaction.

Questions This Demo Answers

How does Nadia actually work under the hood?

Nadia is built from a coordinated system of specialized AI agents rather than a single model. Different agents are responsible for empathy, reflective questioning, tracking behavioral history and patterns across sessions, and an editor-in-chief layer that actively removes biases and hallucinations from responses before they are delivered. All of this coordination happens in a fraction of a second, producing something that feels dynamic, natural, and coach-like — not a generic AI response.

What makes Nadia's approach to questioning different from a standard AI chatbot?

When a user brings a challenge, Nadia's first move is not to provide an answer — it is to ask thoughtful questions designed to build a deeper understanding of the person, the situation, and what kind of help would actually be most useful. This mirrors how skilled human coaches work: understanding before advising. The questions help users surface root causes and clarify their own thinking, so that when solutions emerge they are grounded in a genuine understanding of the individual's specific context rather than generic best practice.

How does Nadia's pattern recognition work — and why does it matter for development?

As a user's conversation history accumulates, Nadia begins identifying recurring themes and behavioral patterns across sessions. If a user has brought up challenges with giving feedback multiple times, Nadia will flag this — not as criticism, but as a developmental observation worth exploring. This longitudinal awareness is one of the most distinctive capabilities of AI coaching compared to one-off human coaching sessions. It creates a developmental thread that grows more valuable the more a user engages, and is one of the key reasons users keep returning to their AI coach.

What does the role-play feature do and how does it work?

Nadia's role-play feature allows users to practice a real upcoming conversation before it happens. You describe the person you need to engage with and the context, and Nadia takes on that persona — responding in character, including with defensive or difficult behaviors you are anticipating. After each message you send, Nadia provides in-the-moment feedback on what worked and what could be improved. At the end she delivers a holistic debrief covering what went well and where to focus. The result is that users arrive at difficult conversations having already practiced them, with both better technique and greater confidence.

Full Demo Transcript

What is Nadia and how does she work under the hood?

Mark: Hey, it's Mark from Valence, and I'm going to show you a little bit about Nadia, our AI-powered leadership coach. The idea behind Nadia is that she can provide always-on, personalized, and actionable guidance for your employees — delivered in the flow of work, in 100-plus languages. Through this technology, you can scale coaching not just to more people, but also in more moments, by providing more frequent, bite-sized moments of learning and development for your leaders and colleagues.

Under the hood, Nadia is made up of different AI agents, which account for the different nuances of the coaching relationship. There are specific agents providing empathy, asking questions, looking at history and patterns of behavior, and ones that act almost like editor-in-chiefs that remove biases and hallucinations. All of this coordination happens in a fraction of a second to deliver something that feels dynamic, conversational, thoughtful, helpful, and ultimately coach-like.

What does it look like to pick up a previous conversation with Nadia?

Mark: I'll put away the PowerPoint and show you how it works. This is a standalone web application, accessible through your phone or computer's web browser. Within this environment, you can see that Nadia has a memory of what we've talked about before to help personalize the advice. She remembers that we were talking about giving tough feedback to a team member, and she wants to bring that back up. I'm going to pick that conversation up.

"Hey Nadia, let's pick that back up. As you might remember, I have someone on my team named Nick who can be defensive when he's feeling criticized. I've got to give some tough feedback to Nick in our one-on-one later today, and I'm hoping you can help me prepare for it. Can you maybe help me think through what to do?"

What you'll notice is that Nadia takes this in and transcribes it perfectly. I'm not prompt engineering my language or speaking formally — I can talk casually. What Nadia does is start to provide what a human coach would do: a little bit of empathy, a little bit of understanding. And then what I think is so unique about our approach — and why we're getting such positive feedback from users — is that Nadia asks you thoughtful questions, trying to gain a deeper understanding of who you are, what the challenge is, and how she can best help you.

She's starting to frame up how we might approach this and what the strategies might be. Through this back and forth, I'll get the chance to reflect on this moment, learn from it, and start to co-create with Nadia how I can best approach this conversation.

What does a more developed coaching session look like — including pattern recognition?

Mark: I'm going to jump ahead to a conversation I've prepared to show you what one of these can look like further along. I'm preparing to give Nick some feedback. Through the back and forth with Nadia, I'm exploring what are the root causes, what's my history with Nick, and how can I give Nadia more information to work with. Eventually, we start to co-create solutions together. Nadia helps me think through what topics I want to cover, what approach or strategy I can use — and this can be customized to be congruent with your organizational values. Eventually, Nadia can even help me come up with a concrete agenda I can use to deliver this feedback to Nick.

Alongside this, different AI agents are running to promote accountability and follow-through. In addition to the coaching, I receive tips and tools to make the behavior stick. I can generate a list of action items to keep myself accountable, email myself a session summary, or schedule a follow-up session with Nadia to debrief how the one-on-one went.

Up in the top right corner, I'll also show you some of the pattern recognition Nadia can do. What this system is recognizing is that I've talked with my coach about challenges delivering feedback before. Nadia is flagging: "Hey, maybe you might be a factor in this dynamic. It's not the focus right now — we're going to help you talk to Nick. But this is something we can come back to as a developmental goal for yourself." That type of pattern recognition and higher-level thinking is one of the reasons why users keep coming back to their AI coach.

How does the role-play feature work?

Mark: One more layer I'll show you is the ability to role-play this difficult conversation with Nadia. If I want to practice this conversation, I can actually have Nadia take on the persona of Nick and respond defensively to me. What you can see is that Nadia responds in character — in this case it's a colleague named Kevin — and gives me guidance on how I can have this conversation, both line-by-line in-the-moment feedback and support, and a holistic overview of what went well, where I can improve, and comprehensive feedback at the end.

The idea throughout all of this is: how can we use these tools to help your leaders and colleagues do their job a little bit better, a little bit faster, and a little bit easier? If you have more questions, please feel free to reach out to anyone from the Valence team. Thank you.