How I mentor and use AI agents
Mentor · Process Designer · 2024 — present
Mentoring is a regular practice for me — and not only in the 1-on-1 format with external mentees. Over my career 30+ people have gone through mentorship: some in my art-director role at MTS, Ozon, and Gazprom Neft, where I systematically grew subordinate leads and strong mid-levels, and some in independent sessions afterward. Most grew: new roles, promotions, moves to dream companies. The biggest time sink was the post-session routine — notes, tasks, tracking agreements. I built an AI agent that automates it. Then a second one, which looks at me as a mentor and suggests techniques for the next session. I published the template for the whole system on GitHub.
Mentoring as a practice
I don't treat mentoring as a "call with advice." For me it's a regular process: a person comes with a request, we work through the situation, capture the takeaways, and return to them in the following meetings.
The format is simple: a one-hour session every two weeks, a shared mentee page with the session history, agreements, and tasks. The pages are password-protected — I don't put mentees' materials out in public, not even in the portfolio. Each mentee sees only their own page.
The main bottleneck is the post-session routine: sorting through notes, laying out the page, writing out tasks, not losing context. An hour to an hour and a half per meeting.
Agent 1 — the post-session routine
The mentee gets a full trace of the session: summary, key topics, decisions, tasks with dates, progress, next steps. All of it is automatically built into the shared mentee page with accumulated agreements and growth history.
Claude Code + Gemini API. Tuned to my mentoring format: which blocks are needed after each meeting, how to capture agreements, how to update the shared mentee page. Ordinary note-taking bots give a five-line summary — here the whole conversation is scanned.
Before, I wrote such a detailed page by hand only for select mentees; there wasn't time for the rest. Now every one of them gets it — on the day of the session, with all the decisions, tasks, and an updated growth history.
Agent 2 — reviewing my sessions
Once the first agent took the routine off my plate, something became visible: reflection can be automated too. I'm a decent mentor, but not a perfect one — the only way to grow is for someone to tell you where you came up short.
The second agent looks at my latest mentoring session and answers two questions: what I did well and what to try in the next meeting. It highlights patterns: gave advice too quickly, held the pause too little, asked a closed question, pulled the conversation onto my own experience where the mentee was still thinking.
At its base is the GROW technique as the framing scaffold. On top of that the agent picks techniques from other sources and links to materials: what to read, where to dig deeper into a specific pattern. There's no hardcoded checklist — the recommendations are live, tailored to each session.
AI here is for reflection. When a model gives you the read, it doesn't weigh on you the way a colleague's opinion does. It's a second you.
It's a mirror that shows me from the outside. I don't take every recommendation at face value. But the simple habit of regular self-review changed the quality of the sessions that followed.
How it looks
Open-source template
Once the system started working for me, I released it as open-source. Names and data are swapped for neutral examples; access privacy is preserved in the template's logic.
Who it's for: a mentor, coach, or team lead who runs regular sessions and doesn't want the follow-up eating their evening.
What's inside: a template for the mentee and session page, a Claude Code config with prompts tuned to the conversation format, a transcription script via the Gemini API, and password-protected access to the pages.
How to get it: clone the repo, plug in your Gemini key, edit the template to your own blocks, and start dropping audio into a folder.
Not a SaaS and not a product. A working template to adapt.
What's next
I want to turn this system into a tool for coaching studios and schools: a single session standard, shared memory per mentee, the reviewer agent as a trainer for junior coaches. That interests me as a product step. I'm looking for partners and people for a pilot — if this resonates, get in touch.
There's no admin panel yet, and I want to finish the mobile version: so it's more convenient to review agreements and tasks. This is the first version of the system.