Your org runs on humans and agents now. Your leadership training covered the humans.
An AI coach built from Marian Kamenistak's published playbooks and 9 live tools, for the questions no leadership course covers yet: managing people alongside AI agents, adoption that actually sticks, delivery flow when half the commits are machine-written. In beta. Human Marian stays one escalation away.
No payment. No card. Two minutes, and you name the price.
The problem nobody's playbook covers
Your best EM has 9 engineers and, since spring, something like 30 agents doing real work in the codebase. Nobody agreed on what an agent may merge without a human reading it. One senior spends his day reviewing agent output and privately calls it a demotion. A junior shipped agent slop to prod, two teams quietly banned the tools, and your CEO is asking why AI adoption is flat.
Delivery didn't speed up. The bottleneck moved: from writing code to deciding what to trust.
And the leadership layer is on its own. The courses your L&D budget pays for teach delegation and feedback for teams that stopped existing this year. When an EM asks "how do I run a performance review for someone whose output is 80% agent-generated?", there is no chapter for that.
Neither does their VP. That's the part nobody says out loud.
What this is
A coach you message the way you'd message a mentor, for exactly this cluster:
- "Two of my seniors refuse to touch agents. They’re my best people. Force it, wait, or route around them?"
- "What’s a fair performance conversation when I can’t tell which output was theirs?"
- "Agent-caused incidents are eating on-call. Who owns an agent’s pager?"
- "The board wants an AI adoption number. Which one won’t blow up in my face in Q3?"
It answers the way Marian answers: a position, a framework, and the move to make on Monday. Not a survey of the internet's opinions.
It's built entirely from Marian's public body of work: his published frameworks, his blog, the 9 live tools in the Engineering Leadership Toolkit, his writing. Nothing from private client sessions. Ever.
It will take wider leadership questions too. But this page sells the one thing nobody else owns.
The obvious question: why not just ask ChatGPT?
Honest answer: for most leadership questions, do. It's good at "how do I run a 1:1" and "how do I delegate". Eight of the nine use cases originally planned for this product got cut for exactly that reason. A free general model already handles them, and selling you a paid wrapper for them would be a scam with better fonts.
What survived the cut is the cluster where a general model fails you three ways.
It has no position.
An average of the internet agrees with whoever’s typing. A coach that never pushes back is a mirror. This one runs on one person’s published, signed positions, and it will tell you when your agent rollout is theater.
There’s barely anything to average.
Mixed human + AI-agent management has almost no written playbook yet. General models thin out exactly where you need them thickest.
It has no next tier.
When you hit the case above any AI’s pay grade, ChatGPT’s answer is another answer. Here, the next tier is a human who grew Mews engineering from 8 to 80 teams through Series C and has 3,400+ sessions behind him. You escalate. That’s the point of the whole ladder.
How the AI and the human fit together
Not a cheaper Marian. A wider one.
The AI coach is the base layer. Every EM, tech lead and staff engineer in the org. The 2am question. The rehearsal before the hard meeting. The fourth "is this normal?" this week that nobody would ever book a session for.
Human Marian is the escalation tier. The reorg call, the VP conflict, the conversation with a name attached. When a thread belongs in a room with a human, it gets flagged, and you book the session.
Today you can afford to give three people a coach at 361 EUR a session. This gives one to forty, and keeps Marian for the calls that actually need him. That's the whole design. Nothing gets replaced. The base gets forty times wider.
Where this actually stands
Straight version, because my name is in the URL.
Exists today: the 9 MCP tools, the published frameworks, the blog, and me. You can use all of that this afternoon, free.
Doesn't exist yet: the coach itself. The conversational layer tying it all together is being built through 2026, early cohort first.
No launch date on this page. I've sat through too many roadmap reviews to publish a date I might miss under my own name. The early list hears first, and hears honestly, including if something slips.
Why the page exists before the product: I build in the order demand proves out, not the order that's fun. If enough of you say "yes, and here's what I'd pay", this gets built faster. If you all say no, you'll have saved me a very expensive year. Either answer is useful. Silence isn't.
Marian
For companies: coach the whole leadership layer, not the top three
Right now your coaching budget covers your three most senior leaders. The other forty EMs, tech leads and staff engineers get a course library and good luck. And they're the ones deciding, team by team, whether your AI transformation is real or a slide.
What changes: every leader in the org gets the daily coaching layer, built on one consistent set of frameworks instead of forty private ChatGPT habits. Your senior three keep human Marian. Hard cases escalate up instead of festering down.
Early-list companies get a call with Marian about their own mixed-team setup before public launch. That call happens whether or not you ever buy.
On the list.
If you asked for the 30 minutes, I come back with times this week. If you didn't, you hear from me before the early cohort opens.
For individuals: the coach your learning budget can't buy
You're the EM whose team went agent-heavy this quarter, and your own manager has less of a clue than you do. A human coach costs more than your entire annual learning budget. Until now the fallback was ChatGPT and hope.
Join the early list and you get first access, early-cohort pricing, and a real vote on what gets built first. The beta is being shaped by the people in it, not shipped at them.
On the list.
You hear from me before anyone else, including the day something slips. No newsletter attached.
Check the work yourself
The claim here isn't "trust me, years of experience". Every source this coach is built from is public today. Go poke at it.
9 live tools you can install right now.
The Engineering Leadership Toolkit is in the official MCP registry: developer and EM salary calculators, a team-lead readiness test, leadership benchmarks, 1:1 playbooks, first-time-manager failure modes, a coaching cost estimator, a mentoring business case builder, and a mentor-vs-coach chooser. marian.coach/mcp
The published frameworks and the blog.
Free, in full, at marian.coach/blog. What the coach says traces back to what Marian wrote.
The track record behind the material.
3,400+ mentoring sessions with 300+ leaders across 17+ countries. Grew Mews engineering from 8 to 80 teams through Series C. Principal Software Architect at Databricks. Engineering leadership at Manta, acquired by IBM.
The community.
2,000+ engineering leaders in ELC. 500+ in the room at the 2026 conference, with speakers from Netflix and Stripe.
What I'm publishing next
The coach is built from what I've written. So here's what I'm writing, in order. Each one lands on the blog first, and goes into the coach after.
- 01
Your best senior refuses to use agents. She's probably right.
Refusal is signal, not resistance. The strongest engineer on your team is usually the first to see the quality cost, because she's the one reviewing the output. Force adoption over her objection and you break the person who was warning you.
- 02
Performance reviews when you can't tell whose work it is.
Output is a team sport now, and some of the team isn't human. Any review that grades volume is grading the agent. What's left to measure is judgment: what they chose to trust, what they caught, what they escalated.
- 03
The AI adoption metric that won't blow up in Q3.
Adoption percentage measures compliance, not value. Rework and revert rate measure whether the work is holding. If AI-assisted work reverts more, you don't have throughput, you have motion.
Tell me where to send them
On the list.
Each post lands in your inbox the day it goes up. Three emails, then nothing until there is something worth sending.
One question, and you can skip it.
What would you pay per month for the coach itself?
Got it. That answer is worth more to me than the signup.
Frequently asked
Is this you replacing yourself?+
No. 1:1 mentoring continues exactly as is, and the hard cases the AI surfaces get escalated to me. The AI is for the forty leaders who were never going to get a coach at all. If it works, I end up in more of the right rooms, not fewer.
What is it built from?+
Public and Marian-owned material only: published frameworks, the blog, the 9 MCP tools, his own writing. No private client sessions, no transcripts, no mentee stories. That's a hard line, not a preference.
What happens to what I type into it?+
I don't sell it, I don't share it, and I don't train anything on it. The coach runs on a model I didn't build, so I'll only use a provider whose terms say they don't train on what passes through, and I'll name that provider before you type a word into it. You're rehearsing hard conversations in there. That only works if it's a vault.
My company pays. Can my manager read my chats?+
No, and I'm not building a way for them to. A company buyer gets seat counts and aggregate usage. There's no screen anywhere that hands them a conversation. Same reason as above: a coach your boss can read isn't a coach.
Why wouldn't I just use ChatGPT?+
For most leadership questions, you should. Eight of nine planned use cases were cut because a free model already covers them. This exists for the one cluster general models average away, plus a human escalation tier ChatGPT will never have.
When can I use it?+
The 9 tools: today, free, at marian.coach/mcp. The coach: 2026, early cohort first. No public date, on purpose. See the status section.
What will it cost?+
I don't have a number for you yet, and I'm not going to invent one to make this page look finished. The form asks what you'd pay. Enough of those answers, and that's the price.
Two minutes, and you name the price
No payment. No card. If enough of you say yes, this gets built faster.