What AI at work looks like today and what it becomes in the next 18 to 36 months 

Most leaders I talk to are somewhere between curious and overwhelmed when it comes to AI. They’ve got Copilot, ChatGPT, or Claude. Maybe Einstein inside Salesforce. Something baked into their HR platform. They’re seeing productivity gains. They’re also fielding pressure from above to show more. 

Here’s what nobody is telling them clearly enough: what they have today and what’s coming next are not a little different. They’re fundamentally different. Not incrementally different. Architecturally different. 

Understanding that distinction might be the most important thing a technology leader can do right now. 

What an agent is 

An agent is purpose-built. It does one thing. It knows one domain. 

Copilot works with you. It aggregates information, surfaces it, helps you move faster. Einstein inside Salesforce does the same thing within that environment. The output is always pointed at you, the user. That’s the design. 

That’s also the limitation. 

Right now, most organizations using AI are using agents exactly this way. Single purpose. Single domain. The value is real. The productivity gains are real. But the agent doesn’t know what the agent next to it is doing. They’re not talking to each other. They’re not building on each other’s output. 

They’re working in parallel, not together. 

Orchestration is what changes everything 

Orchestration is what happens when you connect those agents into a workflow. 

Think about what that means in practice. You create content. An orchestrated system takes that content and automatically builds the email campaign around it. The hand-off isn’t a task on someone’s to-do list. It just happens. Two agents, working together, replacing what used to require two people moving something manually from one platform to another. 

That’s a simple example. Scale it. 

An agent managing your recruiting workflow connects to the agent managing onboarding. Your service desk agent connects to your infrastructure monitoring agent. Your sales intelligence agent connects to your account planning process. Now you’re not just moving faster. You’re operating in a way that wasn’t architecturally possible before. 

And here’s the part most people aren’t talking about yet: when those agents are connected and running, they’re generating something. Every interaction, every decision, every workflow they touch is creating a digital record of how your business thinks and operates. That accumulation of intelligence is not generic. It’s yours. It reflects your processes, your institutional knowledge, your way of making decisions. 

That’s your digital AI IP. And unlike most business assets, you can’t acquire it after the fact. You build it over time, or you don’t have it. The longer your orchestrated systems run, the more valuable and specific it becomes. The harder it becomes for a competitor to replicate. The more embedded it becomes in how your business works. 

The gap that matters 

Here’s what I see when I look at the market honestly. 

Most companies are not ready to make the orchestration leap today. That’s fine. The tooling is still maturing. The use cases are still being defined. Buy versus build decisions are real and complicated right now because every major platform, Salesforce, Workday, Box, ConnectWise, is building agents into its products. And in many cases, buying the purpose-built agent that comes with the platform you already run makes a lot more sense than building something from scratch. 

The mistake would be assuming that because you’re not ready for orchestration today, you don’t need to be thinking about it today. 

Because the gap between where you are and where you’re going is not a technology gap. It’s an architecture gap. And architecture gaps take time to close. The organizations that start thinking about orchestration now, even before they’re ready to execute on it, are the ones who will move faster when the moment comes. 

The ones who wait until the moment arrives will spend years catching up. 

What this means for how you work right now 

You don’t need to abandon what’s working. The agents you’re running today are doing exactly what they’re supposed to do. Use them. Get comfortable with them. Let them generate productivity gains and start teaching your organization what it feels like to work alongside AI. 

But while you’re doing that, start asking a different set of questions: 

• Which workflows in your business are currently split across multiple tools, multiple teams, multiple handoffs? Those are your orchestration candidates.  
• Where is institutional knowledge getting lost between steps?  
• Where are humans acting as connectors between systems that should be talking directly to each other? 

The answers to those questions are not a roadmap for today. They’re a roadmap for 18 to 36 months from now. And the organizations that have that roadmap, even roughly sketched, will move faster than the ones starting from zero when the technology catches up to the ambition.  

The question worth pressure-testing 

If your current AI tools started talking to each other tomorrow, would your infrastructure hold the seams? 

Not the tools. The infrastructure underneath them. The architecture that would need to connect them, scale with them, and protect the IP they generate. 

If the honest answer is “I’m not sure,” that’s not a reason to panic. It’s a reason to start asking better questions. Which is exactly what part three of this series will be about. 

This is part two of a four-part series on AI, infrastructure (on-premises and cloud), and the decisions that matter most right now. Missed Part 1: We’ve Been Here Before.