Technology Decisions are Leadership Decisions – Not IT Decisions

Author

Charlene Li
November 10, 2025

Out: Multi-year AI implementation roadmaps crafted by IT. 

In: Modular AI implementation initiatives powered by executive leadership. 

Here's the uncomfortable truth my “Winning with AI” co-author, Katia Walsh, and I keep running into: So many leaders look at AI as another tech project: build it slowly, and build it to last. In the old days, once a system was set up, it was there to stay. 

AI doesn’t play by those rules. The technology is evolving too fast, and the cathedral approach that served us well for decades is now our biggest liability.

So how do you build something when you know it's going to have to change?

Technology Decisions Are Leadership Decisions

Let's start with something fundamental that too many leaders are getting wrong: AI is not an IT decision.

Katia and I have worked with leaders for decades, and we keep seeing the same pattern. Someone asks, "What's our AI strategy?" and the immediate response is, "Let's have IT figure that out."

Here’s what leaders should know: AI should not be the first word in your strategy. Having an ‘AI strategy’ makes about as much sense as having a ‘telephone strategy:’

💡 AI is a technology to accomplish your business strategy, not the strategy itself.

As leaders, we provide the why. Your engineering teams bring the how. That means you need to understand enough about AI platforms to explain what you're trying to accomplish. You don't need to code, but you do need to speak the language well enough to have real conversations with your IT teams, your vendors, and your partners.

From Cathedrals to Legos: A New Building Philosophy

Traditional IT projects are like building cathedrals. They’re beautiful, well-planned, permanent structures with multi-year roadmaps. And in the age of AI, they're going to be obsolete by the time they're done.

I'm not exaggerating. Look at what's happening right now. Model Context Protocols. Agentic Processing Automation (APA). Multi-agent systems built on LangGraph, Autogen, and CrewAI. These technologies are being built as we're trying to implement them.

The solution? Build with Legos instead.

Think modular. Think adaptable. Think about swapping pieces of your technology platform as better options become available—without having to dismantle everything and start from scratch.

There's precedent for this. 

Back in 2002, Jeff Bezos wrote an infamous memo mandating that Amazon would use APIs for everything. No more shipping data from one place to another. That memo completely changed how they operated, making everything more flexible and modular.

The Four Layers of AI Platforms

In our forthcoming book, “Winning With AI,” my co-author Katia Walsh and I use a kitchen analogy to break this down. Your AI platform has four distinct layers, and you should be able to swap components within each layer as technology improves:

Layer 1: Data (Your Ingredients) 

It’s true that better ingredients make better food, and better data will produce better results. But the most important thing is just to get started.  

Your data will always be messy. You're constantly generating new data, and discovering new types of data. Stop waiting for perfection and get cooking. 

Deploy AI with the data you have and work in parallel, not linearly. AI itself can help make sense of messy, incomplete, unstructured data.

Layer 2: AI Models (Your Equipment) 

You wouldn't use a microwave to bake a cake. You need different tools for different jobs.

You don't need to build your own models. Use powerful existing models and connect them to your specific data through tools like retrieval augmented generation. When your data updates, AI instantly has access—no retraining required.

Layer 3: Orchestration (Your Head Chef) 

This is where things get really interesting. You have data over here, models over there…but who's making sure everything works together?

McKinsey built an internal AI platform called Lilli that now has 25,000 internal AI agents. Here's what that looks like in action: they could have five consultants build a sales performance prediction model over twelve weeks, or have their AI agents do the same work in three hours, more accurately.

A single model can't do complex work. But orchestrated agents working together can change everything.

Layer 4: Consumption (Your Service and Presentation) 

Here's a stat that should terrify you: 70% of AI projects fail simply because people aren't using them.

This isn't a technology problem, it's an interface problem. If you have poor interfaces, AI gets confined to a small number of technical users. 

The best AI disappears into existing workflows. Think Microsoft Copilot embedded in the tools you already use. Think Salesforce Agent Force built into your CRM. 

There are three stages of better consumption:

  1. Embedded: AI in the apps you already use

  2. Self-service: Non-technical users building their own solutions without IT support

  3. Ambient: AI that disappears completely: voice, vision, gestures, naturally in the flow of work

What Leaders Need to Do Now

Challenge your IT teams with these questions:

  • How are we building versus assembling?

  • What's the state of each layer? Our data, models, orchestration, and interfaces?

  • What's our plan to swap these pieces as technology improves? 

  • Do we have the APIs and hooks to make modularity actually happen?

The Bottom Line

Technology is a leadership decision. Get your hands dirty. Get into the messy business of understanding AI platforms. Don't wait for perfect data. Don't build your own models. Make it modular so you can adapt quickly.

💬 Your Turn

What are your biggest challenges when it comes to AI leadership right now? How is your organization approaching the need for modular AI tools?

What I Can’t Stop Talking About

  • AI is coming for some jobs. It’s up to leaders to support their employees through career changes with reskilling and assistance finding new roles. This may be a rocky moment, but you don’t have to reinvent the wheel to get through it. 

  • “Doing AI” takes more than you might think. Simply implementing a few tools isn’t transformation: it’s tech stack bloat. Here’s what a full AI integration actually looks like. 

  • Want to learn more about the alphabet soup of agentic AI, like MCP, APA, and LangGen? Join me for my next livestream on Tues. Nov 11th where I’ll be discussing what leaders need to know and ask about agentic AI.

 My Upcoming Appearances/Travel

If you found this note helpful, please forward and share it with someone who needs the inspiration today. If you were forwarded this, please consider subscribing.

Join me on Facebook, Instagram, LinkedIn, and Twitter.