AI Coaching Platforms: What to Automate, What Stays Human

The hour a coach spends with a client is the one part of the week AI should leave alone. Almost everything around that hour is fair game.

Coaching platforms are hearing the opposite. A competitor announces an "AI coach" that chats with clients, and the roadmap fills up with a bot that gives advice in the coach's name.

Look at where a coach's week actually goes. The sessions are what clients pay for. Around them sit the notes after each call and the follow-up that should have gone out yesterday. There's the course module that's been almost done for a month. There's the Tuesday post that still has no topic. None of that needs the coach's judgment, and all of it eats the coach's time.

We've worked with Milana Leshinsky on Coaching Genie, her coaching platform, for five years. The AI we built into it lives in that second pile. Session notes are AI-assisted. Helping coaches write course content is fully automated, and so is the research behind content ideas.

This is the split we'd use for any coaching product, and the reasoning behind it.

One test for every feature

Before a coaching feature gets AI, ask one question: would the client feel short-changed if they knew software did this part?

Nobody minds that a transcript was typed by a machine. Everyone minds if the advice on their career change came from a model while the coach was at lunch. That test sorts almost every idea on a coaching roadmap, and it sorts them fast.

Start with session notes

After a session the coach has to remember what the client committed to and what to raise next time. Done by hand, it's a block of typing after every call, or it quietly doesn't happen.

An AI version drafts the note from the session recording or transcript: a summary and the client's commitments, with anything left open flagged for next time. The coach edits it and saves it. Coaching Genie's notes are AI-assisted, and "assisted" is the word that matters. A model summarizing a conversation will now and then put a commitment in the client's mouth that they never made. The coach's review is what keeps the client's history true.

Then the prep before each session

Once notes are structured, the next step is cheap. Before each session, show the coach one screen with what the client committed to last time and what has happened since. It's reading, not judgment, which makes it a good job for a model whose only source is the coach's own notes.

Course content, drafted from the coach's own material

Most coaches sell a program, and the program is where their weeks disappear. They know the material cold. Turning it into modules and lessons with a consistent structure is the slow part.

This is where we went furthest with Coaching Genie: helping coaches write course content is fully automated. Features like this tend to follow the same pattern. The coach supplies the raw material, meaning their method and outline or recordings of how they teach it. The system turns that into a structured draft, module by module, in the coach's own words. Code checks that every lesson has the parts the platform needs before anything is saved. I'd always leave the publish button with the coach.

Content-idea research can run on its own

Coaches are told to post every week, and the research is the part nobody enjoys: working out what the audience keeps asking about and checking it against what's already been covered. We fully automated content-idea research for Coaching Genie and for Xperiencify. It was a manual workflow, and AI automation replaced it.

It automates well for a simple reason. The output is a list of options a person picks from, so a weak idea costs a few seconds of reading, not a bad post.

Sort your coaching roadmap

Paste your AI ideas. ChatGPT will sort them into what to automate and what to leave with the coach, and suggest which one to build first.

“|”

The advice stays with the coach

A coach is paid for judgment about one person's situation. A model can explain a framework. It can't know that this particular client says "I'm fine" right before they go quiet for a month, because the coach learned that over many calls. Advice in the coach's name has to come from the coach.

There is a middle ground that works, and it's practice. Irresistible Bot, the AI copywriting agent we built for Vrinda Normand, supports her students around the clock in her voice and guides them through her copywriting frameworks. The job there is applying a known framework to the student's own copy. That's a very different thing from deciding something about a client's life.

So does accountability

Clients show up because a person expects them to. Automated reminders help at the margins. An AI "accountability partner" is a nudge that clients learn to ignore. Let a model draft the follow-up if you like, but it should go out from the coach, reviewed by the coach.

Anything close to a crisis goes to a person

Coaching isn't therapy, but clients bring real problems into it. A model shouldn't be the first to answer a message from a client in distress. Route it to the coach, and don't auto-reply.

Where coaching AI goes wrong

The failures are predictable. Notes that invent commitments are the first, and the coach's review step catches them. Course drafts that flatten every coach into the same voice come next. That one is a context problem: drafts need the coach's own material and writing samples, the same way brand-grounded AI copy does.

Then privacy. Session recordings are the most sensitive data a coaching platform holds. Get consent to record, and set a retention rule for the recordings. Check what your model provider does with what you send. Above all, make sure one coach's clients can never surface in another coach's workspace. That's gap 1 in our AI production checklist, and a coaching platform can't afford to learn it from a client.

Simplicity is the constraint

Here's how Milana describes what Coaching Genie's users value: "What people love most about it is the simplicity, how simple it is and how easy to understand the tool is."

Every AI feature has to clear that bar. A good one removes a step: the note is already drafted, the module already has its structure. A bad one adds a chat window and a setting to explain. If an AI feature needs a tutorial, it failed before launch.

Adding AI to a coaching platform?

We build the AI layer into existing coaching products as a flat-price engagement: one senior engineer, your repo and your cloud account from day one, usually two to five weeks. Send your email and we'll set up a 30-minute call within 24 hours.

The coaching AI check

A feature should pass all seven before it ships.

Would clients be comfortable knowing software did this part?
Does a coach review the output before a client sees it or before it enters a client's history?
Does the model work from the coach's own material rather than generic advice?
Does it remove a step for the coach instead of adding a screen?
Do you have consent to record sessions, and a retention rule for the recordings?
Is it impossible for one coach's client data to appear in another coach's workspace?
Does a message from a client in distress reach a person, not a bot?

Where I'd start

If I were starting on a coaching platform tomorrow, I'd build AI session notes first. They happen after every call and the coach checks every one, so quality problems surface fast. They also feed everything after them, from prep screens to follow-ups. Course-content drafting comes next, because that's where coaches tend to lose weeks. The client-facing chatbot waits until both are working, and it may never need to exist.

The AI Coaching Stack: What to Automate and What Must Stay Human