Map Out Your Workflows

By Lydia Chilton

Before you build an AI strategy, you have to understand how you, your team, or your whole organization are getting work done.

In most workplaces, knowledge about how things get done is buried in email threads, Slack messages, and informal agreements. When information is scattered and inconsistent, it’s nearly impossible for AI to take on the drudgery that underpins complex operations.

That’s why the first step is to understand how work gets done — by you, your team, or your entire organization. Especially the mundane stuff. I’m talking about filling out forms, chasing stalled conversations, and shepherding documents through layers of approval and revision.

AI agents aren’t just suggesting text anymore. I’ve seen agents handle everything from simple scheduling to more complex coordination tasks, like planning resource allocation or accounting for unexpected hiccups that could send disruptive ripples across teams. There’s more upside — and stronger competitive pressure — every day. 

But agents can only help if your systems make sense to machines. That means defining inputs and outputs clearly, writing down rules, digitizing steps, and building APIs, which let digital tools share information automatically. 

Here’s how I suggest starting:

• Define your workflows. What tasks need to get done? What information is needed to do those tasks? What does success look like?
• Remove manual blockers. If someone has to write information by hand or transfer it manually, redesign the process.
• Standardize context. Agents shouldn’t have to guess what “done” means. Make rules and data explicit.
• Let agents learn. If you show them a few examples, they’ll remember what to do—and know when to ask for help.

Personally, I’m excited for a future where we don’t waste time filling out forms or navigating clunky systems. The less time we spend pushing paper, the more time we can spend doing meaningful work. That’s the real promise of this technology.

Tomorrow, my colleague Zhou (Jo) Yu will share her perspective on why starting with the right use case is key to deploying AI agents in your organizations.

If you’d like to learn more about partnering with Columbia researchers working at the forefront of applied research in AI, visit the DAPLab website or contact Lydia Chilton

Learn More

Get in Touch

Image Lydia Chilton Associate professor of computer science at Columbia Engineering, a visiting scientist at Amazon, and a Data Science Institute affiliate