AI Coaching Agents: Two Ways to Get One, and the RFP Terms to Ask For

The phrase is turning up in RFPs and vendor decks now: “AI coaching agent for managers.” If it has landed on your desk, you’ve probably noticed that nobody defines it. One vendor uses it for a chatbot. Another uses it for a whole platform with assessments and dashboards. A third means something that listens to sales calls. You’re expected to buy one, and it isn’t clear what one is.

This guide explains what an AI coaching agent is, from the side of the people who build them. It also covers the two ways to get one. You can run a coaching agent on your own systems, or you can buy it inside a coaching platform. Both are real options, and the RFP terms you need are mostly the same either way.

Before going further, you should know where we stand. Culturro builds HR agents for companies. We also own Risely, an AI coaching platform for managers. So we sit on both sides of the choice this post describes. When we use Risely as an example, read it as our own product, not as an independent review. We’ll say plainly where something is our judgment rather than a published finding.

The three things “AI coaching agent” can mean

Search for the phrase and you’ll find three quite different things under the same name.

The first is about training an AI support bot. Here “coaching” means teaching the bot to answer customer questions better, and the “agent” is the bot itself. No person is being coached at all.

The second is coaching for contact-center staff. Software listens to calls, scores them and suggests what a rep could have said. The “agent” is the human call-center agent, and the AI is the coach. This is a large market, and a lot of the statistics you’ll see in vendor decks come from it.

The third is coaching for managers. A manager has a real situation at work and talks it through with an AI that helps them decide what to do. This is the one HR buyers mean, and it’s the only one this post covers.

It’s worth checking which of the three a vendor is selling before you read their numbers. A figure about call scores tells you nothing about whether a new manager will handle a hard conversation better.

What an AI coaching agent is

An agent, in plain terms, is software that does a defined job for a person. It takes what that person brings to it and works through the problem with them. At the end it produces something useful. A coaching agent’s job is narrow on purpose. On our HR Agent Map, we define the Manager Coaching Agent as an agent that helps a manager handle the situation in front of them.

That means the problem the manager has today, not a course they might take next quarter. A report may have gone quiet in meetings. Two peers may have stopped talking to each other. A promotion request may be one the manager can’t grant, or a deadline may have slipped and now needs a conversation. These are the moments when most managers have nobody to ask.

Here’s what that looks like in practice. This is a composite made up for this guide, and the details are invented. Maya runs a team of six in finance operations. One of her strongest analysts has asked for a promotion, and the budget for it isn’t there this year. Maya opens the coaching agent in Teams on Monday morning. It doesn’t hand her a script straight away. It asks what the analyst actually wants. Is the issue the title, the pay or the work? It also asks what Maya can offer instead. Next it offers a couple of ways to frame the conversation, and together they settle on an approach and the first few sentences. On Thursday the agent checks in to ask how it went.

That example shows each part of the agent’s job. It sits in Slack or Teams, where the manager already works. It asks the questions a good coach would ask and offers frameworks and options. Finally it helps the manager settle on an approach and the actual words. What it produces is a private conversation that ends in a concrete plan, plus optional follow-ups that check how things went. The manager also stays in charge. The agent suggests, but the manager chooses what to do and owns the result.

On the HR Agent Map we rate this agent as proven, which means it’s widely deployed and well understood. Its data sensitivity is medium, because managers will type real things about real people. It’s also the first agent in our Manager Enablement sequence, and the reason is simple. It needs only chat access and an agreed privacy model to start.

Two ways to get one

A coaching platform and a coaching agent aren’t competing kinds of product. A platform contains a coaching agent, and then adds other things around it. So the real question isn’t “agent or platform?” It’s how you want to get the agent. You can run one on your own systems, or you can buy one inside a platform.

The table sets the two routes side by side. We’ve added a human coach as a reference point, since most HR teams are comparing against one.

	Agent on your own systems	Agent inside a platform	Human coach
Where it lives	Your Slack or Teams, on your company’s accounts	The vendor’s product, often reached through Slack or Teams	Calls or meetings booked with the coach
What it knows	Your coaching framework, values and policies, which you supply	The vendor’s skills framework and content, plus whatever you can add	The coach’s training and what the manager tells them
Who owns the data and the output	You, under your own privacy rules	The vendor holds it, under the contract you sign	The coach holds their own notes, under their professional ethics
What you configure	Almost everything, from the framework to the privacy model	Settings the vendor exposes	The brief you give the coach
What comes with it	Only what you build	Often assessments, a content library and reporting	Judgment and a human relationship
Who it reaches	Every manager, any time of day	Every licensed manager, any time of day	Usually a few senior people

Read the table one row at a time and ask which answer your company needs. The “what it knows” row matters most for quality. A coaching agent only gives good advice if it’s grounded in a framework and in the way your company actually works. If your values and policies are well written down, running the agent on your own systems lets it use them fully. If they aren’t, a platform’s built-in framework gives you a sensible starting point.

The “what comes with it” row is where platforms earn their price. If you want skill assessments, a content library and reporting for HR, a platform gives you those on day one. Building them yourself is real work, and most HR teams shouldn’t.

Where Risely fits

Risely is our example of the platform route. Again, it’s our own product. Its coach is called Merlin, and Merlin coaches managers natively inside Slack and Teams. These are the figures Risely publishes about its own use, as of October 2026. 87% of invited users engage in their first week, and 82%+ are still engaging at day 30. 86% say they’d continue coaching with Merlin. Users have an average of 4.5 coaching conversations a month, with about 30 messages in each. They see an average 26% skill improvement in 12 weeks. More than 5,000 users have been coached across 40+ organizations. Risely covers 83 skills and supports 40 languages. It takes about 5 minutes to get from signup to a first coaching session.

These are Risely’s own numbers, not an independent study. They’re useful mainly as a benchmark. When another vendor quotes engagement, you can ask how their week-one and day-30 numbers compare. On price, Risely’s site puts traditional coaching at $500 an hour or $25,000+ a year. That cost is why the human coach in the table usually reaches only a few senior people. You can read more about how we built Risely on our Risely work page.

Why this agent needs chat access and no HRIS data

Many HR AI projects stall on data. The HRIS is messy and performance records are patchy, so nobody wants to connect them. A coaching agent avoids most of that. All it needs is a coaching and skills framework and your company’s values and policies, plus a connection to chat. It needs nothing from the HRIS or from performance records, which is where so many other projects get stuck. If you’ve read our post on HR data readiness, this is why a coaching agent is often the easiest first agent to start with.

Privacy: who sees the coaching conversation

This is the question that decides whether managers will use the agent at all. A manager who thinks HR can read their messages will type polite, safe things. That makes the coaching useless. So before features, you need to know who can see what.

On either route, you’ll meet three privacy models:

• Private to the manager. Only the manager sees the conversation. HR may see aggregate trends, such as which skills managers are working on across the company.
• Summaries only. The manager’s own manager or HR sees a short summary, but not what the manager typed.
• Visible to HR. HR can read the conversations themselves.

Our definition of the Manager Coaching Agent uses the first model. Conversations are private to the manager. They’re never shared with the manager’s own manager or with HR, and they’re not kept as HR records. Only aggregate capability trends leave the system. We think this is the only model that gets honest use from managers. We don’t have a study that proves it, so treat it as our view.

Risely’s published stance is a working example of a mixed approach. When a manager uses Merlin on their own, the coaching is private. When the company assigns coaching to a manager, it shares summaries only. That split makes sense. If HR has asked someone to work on a skill, a summary of progress is fair. If the manager came on their own, it’s their space.

Whichever route you take, put these questions in the RFP. Who can read a conversation, and who can read a summary? Is a conversation ever kept as an HR record? Can a manager see what’s shared about them before it’s shared? What happens to the data when the contract ends? Our post on AI governance in HR covers how to set these rules before you choose a tool.

What an AI coaching agent can’t do

An agent is good at the everyday situations a manager faces many times a year. It’s available at 9pm before a hard conversation, and it doesn’t get tired of the same question. But it has real limits, and you should hear them before a vendor’s demo.

A human coach is still better for senior or high-stakes personal work. A new executive working through how they lead, or someone dealing with a career crisis, needs a person. They need someone who can read the room, hold a long relationship and challenge them with care. An agent won’t replace that, and we wouldn’t sell it as if it could.

An agent is also only as good as what it’s grounded in. Without a coaching framework and your company’s policies, it gives generic advice. Generic advice can be wrong for your company, for example on how promotions or pay conversations really work there.

People also push back on feedback from an AI. A study by Xueming Luo and colleagues in the Journal of Marketing looked at AI coaches for sales staff at two fintech companies. It found that AI coaches are strong on data but lack a human’s interpersonal skill in delivering feedback. So the sales staff pushed back on feedback from the AI, and the top-ranked staff pushed back the most. You can read the American Marketing Association’s summary of the study. The researchers then tried a mix. The AI produced the feedback, and human managers delivered it. The test used the lowest- and highest-ranked staff, and for both groups the mix beat the AI coach alone and the human coach alone. The study was about sales, not managers, so treat it as a signal rather than proof for your company. Other researchers are building on the same idea. One example is Socratic, an AI system presented at AAMAS 2025. Its authors describe it as complementing human coaches by guiding teams in real time while they work. It was tested in two experiments with pairs of people, not with managers at work. The practical lesson for HR is that an agent works best for preparation and practice. It shouldn’t be the only voice a manager hears about their performance.

Choosing a route

My advice is to start from what you want around the agent, not from the agent itself.

Buy inside a platform when you want things you wouldn’t build yourself. That means skill assessments, a content library, reporting for HR and a product that’s ready in minutes. Most companies that want manager coaching quickly, without an internal project, should take this route. If you’re comparing platforms, Risely’s buyer’s guide to AI coaching platforms sets out how they differ. It’s written by us, about a market our own product is in, so read it with that in mind.

Run the agent on your own systems when it should use your company’s existing tools, policies and data. That fits when you already have a clear leadership framework and strong views on privacy. It also fits when coaching is meant to be one agent among several, on systems you control. Our post on HR AI software versus custom agents goes through that decision in more depth.

Neither route is wrong. The mistake is buying without knowing which one you’re on. If you don’t know, a platform’s reporting becomes a surprise privacy problem. Or an own-systems build quietly turns into a content library project.

Terms to put in the AI coaching agent RFP

These terms work for either route. Each one comes from a section above.

1. The agent coaches managers on their own work situations, not call scores or bot training.
2. It runs in the chat tool your managers already use, which is Slack or Teams for most companies.
3. It’s grounded in your coaching framework, values and policies, and you can update them.
4. The manager decides what to do, and the agent doesn’t act on their behalf.
5. The privacy model is stated in writing: who sees conversations, who sees summaries, and what is aggregated.
6. Coaching conversations are not kept as HR records.
7. No HRIS or performance data is needed to run it.
8. Follow-ups after a conversation are optional and controlled by the manager.
9. The vendor shares week-one and day-30 engagement, with how each is measured.
10. It’s clear what happens to the data when the contract ends.

Talk it through with us

If you’d like to see what a coaching agent would look like on your own systems, book a 30-minute call. We’ll go through your framework, your chat tools and the privacy model your managers would trust. You can also look at the HR Agent Map first, to see where the Manager Coaching Agent sits among the other HR agents we build.