How AI Actually Lands: Top-Down and Bottom-Up

Adapted from two keynotes I gave in the past week — the 2nd Bay Area Young Entrepreneurs Day, and the Lingnan College (SYSU) Hong Kong Alumni Annual Gala.

I keep hearing the same self-consolation in founder groups:

We’re embracing AI too. Bought a system, brought in a consultant for a talk.

That’s not embracing AI — that’s glancing at AI.

Whether a company is actually moving has nothing to do with that. It has to do with two things: is anyone at the top setting the tone, and is anyone at the bottom actually running with it. Miss either side and AI will never grow into the flesh of the company.

1. Top-down: AI is a CEO job

Whether AI lands in your team is 90% about the person at the top.

Not the CTO. Not the digital-transformation department. Not the AI consultant. You.

The clearest reference is Shopify. In April 2025, CEO Tobi Lütke sent an all-hands memo. The team thought it was “too harsh” and was about to leak it — so he just published it himself on Twitter. The core line:

Reflexive AI usage is now a baseline expectation at Shopify.

Reflexive — knee-jerk; whenever something comes up, the first thought is can AI do this? That’s the baseline expectation for every Shopify employee. The hard rules followed: before hiring anyone new, you have to demonstrate that AI can’t do the job. And “how well you use AI” went directly into the 360° performance reviews.

Shopify is a $100B company with eight thousand people, and the CEO is calling the shot personally. Companies an order of magnitude smaller than that ought to be more ruthless about this, not less.

So what can a boss actually do tonight? Three things.

First, use it yourself, every day. Take tomorrow’s hardest deal memo and think it through with AI in front of your team. If the boss doesn’t model it, nobody else will really use it.

Second, write it into the management system. Explicitly endorse AI tool usage. Explicitly state that “people who can’t keep up with AI will fall behind on performance and be cut.” Process beats slogans, every time.

Third, rewards and consequences, clearly. Bonuses and case-study spotlights for people using it well. Reminders, reassignments, and exits for the stubborn.

2. Bottom-up: do one thing — reimburse

Once the tone is set, stop doing the next thing. No AI committee. No standardized tool selection. No RFP.

The one thing the company should do is send an email:

Starting next month, every employee gets ¥500 per month for AI tools, freely reimbursable. No approval required, no report needed. Submit the receipts at month-end.

That’s the entire email. Three months later, you’ll find Finance teaching itself AI reconciliation, HR using AI to screen resumes, the investment team running its own AI sector maps — none of which you planned.

Employees know best where they’re stuck. Hand them the tools, give them room, then step out of the way.

Why this way? Two reasons.

First, the front line understands the problem. The essence of AI is solving problems, and to solve a problem, you first have to understand it. What the boss sees are outcomes — sales down, project slipping, margin missing target. These aren’t single problems; AI can’t act on them directly. What the front line sees is the situation: how to phrase a follow-up so the client actually replies, which item on the due-diligence list is the load-bearing one, how to finish 100 reconciliations today. These, AI can solve directly.

Put the people who understand the problem in front of the strongest problem-solving tool — this is common sense, not innovation.

Second, AI’s value emerges in the capillaries. Strategic-layer KPIs are revenue, profit, market share — at this layer, AI becomes a question of company culture. Capillary-layer KPIs are “20 customers this week, 99% reconciled this month, 30 pieces of content this quarter” — at this layer, AI grows into the flesh of the company.

Culture is the skeleton; the capillaries are flesh and blood. Lose either side and the company can’t stand up.

Closing

AI is strategy — it has to travel top-down. AI is a way of working — it has to emerge bottom-up.

The top sets the tone; everyone else runs.

Tone without freedom: the company becomes another PowerPoint “AI transformation project.” Freedom without tone: the company splinters into small workshops, each going its own way.

AI doesn’t land on its own. It needs you to set the tone, and it needs you to step aside.