The Room Got Quiet. That Was the Point.
I sat in the front row at a YPO Retail CEO Summit in Montreal and watched a room of 158 experienced operators go silent.
The trigger was a single statistic from Verne Harnish: a SaaS company had replaced the output of 45 engineers with 5 A-level engineers plus AI — and delivered three times more releases at a fraction of the cost. The room didn’t cheer. It went quiet. Because everyone was doing the same mental math: What does this mean for my payroll line? My org chart? My company?
That silence was worth more than most keynotes I’ve attended. It meant the abstraction was finally gone.
This week, across a YPO summit in Montreal, a CEO hands-on AI workshop in Vienna with 54 leaders, a session at DHBW Lörrach with students, and my own founder discipline around what I choose to publish — the same question kept surfacing in different forms:
When AI can do the thinking, the writing, the building, and the analysis — what exactly is left for us to protect?
I want to give you my honest answer. Not a framework I found online. One I am actually living right now.
AI Is Already Inside Your Operating Model. The Question Is Whether You Know It.
Harnish introduced what I think is the defining metric for this era: Return on Payroll. Gross margin dollars divided by total payroll. Benchmark it against where you were in 2019. If that ratio has deteriorated and you haven’t noticed, you have a structural problem hiding inside a familiar spreadsheet.
This is not a future-of-work conversation anymore. It is a present-of-operations conversation.
In Vienna, 54 CEOs spent a full day not listening to another AI trend report — but actually building. Claude Code. Agents. Audit loops. Real company workflows. Hands on keyboard. One CEO said, after completing something that used to require a consultant and three internal meetings: “That took me eight minutes.”
That moment — that specific eight-minute moment — is when AI stops being abstract for a leader. And until that moment happens, you are still operating on borrowed intuition.
The most dangerous sentence I hear from Mittelstand and SME leaders right now is still: “Our IT team is looking into it.” That is not a strategy. That is a delay that compounds.
AI fluency is now a CEO responsibility. Not because every operator needs to write code. But because you cannot govern, price, or lead what you do not understand at a working level.
The Three-Layer Problem: Logos, Pathos, Ethos
Harnish framed it cleanly, and it has stayed with me all week.
AI will outperform us on logos — logic, analysis, calculation, pattern recognition. It already does, in most domains, at scale and speed we cannot match.
AI is becoming frighteningly capable at pathos — language, emotion, persuasion, narrative. If you think your writing, your emails, your reports, or your pitch decks are safe from automation, spend thirty minutes with a well-prompted model and reconsider.
But ethos — trust, character, lived commitment, founder-level standards, the specific soul of what you are building — that cannot be automated. Not because machines lack the words for it. Because ethos is not words. It is the accumulation of decisions made under pressure, over time, in full view of the people who depend on you.
Harnish’s line was simple: “Don’t delegate the soul.”
I’ve been thinking about that ever since, because I’ve seen what the alternative looks like. Companies that automate the obvious work, then the harder work, then the uncomfortable work — and wake up one day efficient but strangely empty. No edge. No taste. No reason to exist beyond margin.
Apple still has it. Amazon still has it. A 19-location family ice cream brand in Texas still has it. Some much larger brands are actively fighting to recover it. That recovery is expensive and slow.
The Ego Filter: Why I Am Posting Less Than I Know
This week I also wrote about something more personal: the discipline of not publishing.
LinkedIn trains founders to share early, share often, build in public, turn every insight into content. Sometimes that is genuinely generous. Sometimes it is performance dressed as transparency.
I am currently building in a domain — an AI-first longevity platform — where a sentence can imply a claim, a claim creates expectation, and expectation moves faster than evidence. And AI makes this worse, not better. A model is excellent at making unfinished thinking sound mature. It can turn uncertainty into a framework, a hypothesis into a conviction, a weak signal into a confident paragraph.
So I have been building a different discipline: share the thinking, not the protected substance. Share the method, not the premature claim. Share the founder discipline, not the scientific detail before it is ready.
I call this the Human-in-the-Loop Ego Filter. It asks one question before you hit publish:
Are you posting because this is useful — or because visibility feels good?
That question hurts. Which is usually why it matters.
The same principle applies inside companies. The human-in-the-loop is not only a quality control mechanism for AI outputs. It is a leadership discipline. It is asking: Is this decision ready? Is this claim grounded? Is this process actually working, or does it just look efficient in a slide deck?
The Limit Is No Longer the Tool
At DHBW Lörrach, I watched a student take a folder of lecture materials and, inside one afternoon, build a fully interactive learning dashboard — topic sections, exam relevance mapping, practice questions, flashcards, a 14-day learning plan, podcast-style summaries, and visual elements. No complex app stack. No IT project. No three-month rollout.
The tool was not the constraint. The constraint was knowing what good looks like and being willing to define it.
I said something to the students that I also believe applies to every SME and Mittelstand operator I work with:
The limit is no longer the tool. The limit is whether you know what you want, whether you can define what “good” looks like, and whether you are willing to get your hands dirty.
This is what I mean by AI meets EQ in practice. Emotional intelligence is not the soft counterpart to technical skill. It is the operating layer that makes technical capability usable. Knowing your own standards. Knowing your team’s limits. Knowing where your company’s soul lives — and what you will not trade away for efficiency.
The Framework: Three Things That Must Stay Human
Across this week — Montreal, Vienna, Lörrach, and my own reflection — I keep returning to the same architecture. I call it the Non-Delegable Stack:
1. Standard-setting. AI cannot define what excellence looks like for your company. It can execute against a definition you give it. The standard is yours.
2. Trust architecture. Governance, relationships, accountability, and the culture of a company cannot be automated. They are built through presence, consistency, and the decisions you make when no one is watching.
3. Soul documentation. This is the one Harnish assigned to the room, and I am taking it seriously: write down, on one page, the soul of what you are building — so you know exactly what must stay human when automation becomes convenient.
If you do not do this deliberately, the default is drift. Slow, comfortable, margin-friendly drift — until the company is efficient and empty.
Five Things to Do Next Week
-
Calculate your Return on Payroll. Gross margin divided by total payroll. Compare it to your 2019 or 2022 baseline. If the ratio has moved significantly and you don’t have an explanation, that is your first signal.
-
Spend eight minutes building something with AI. Not reading about it. Not delegating it. Open Claude or ChatGPT, take a real process from your company, and build a working draft of a workflow. The eight-minute moment changes your baseline.
-
Write the soul of your company on one page. Not your mission statement. The actual answer to: What would we never automate, even if we could? What would make us unrecognizable if we lost it?
-
Run the ego filter on your next three communications. Before you send the strategy update, the client proposal, or the team memo — ask: Is this ready? Or does it just sound ready because AI helped me write it?
-
Identify your highest-leverage Shadow AI risk. If you have not given your team safe, governed AI tools, they are already using ungoverned ones. That is not a compliance problem. It is a leadership gap. Address it this week.
I come back from this week with one clear conviction: the leaders and companies that will thrive in the AI era are not the ones who automate the most. They are the ones who are most deliberate about what they protect.
Efficiency is table stakes. Soul is the edge.
Where in your company are you most tempted to delegate something that should stay human — and what would you lose if you did?