There's one thing I've noticed about how people make decisions when they're looking to purchase new software.
They ask for a demo first, before thinking about what problem they're actually trying to solve.
It's almost like the software ends up driving the decision.
I think that's the wrong way to approach agents.
If you start with, "What can this agent do?" you'll probably end up in an endless loop.
Agents can do a lot. They can interact with software, retrieve information, make decisions, call APIs, move data around, and execute tasks.
And that's exactly the problem.
Without a well defined use case, you'll keep finding more things the agent could potentially do.
So step back first.
Define the problem you're trying to solve.
Then map out the workflow. What needs to happen? What decisions need to be made? What systems need to be accessed? Where are the repetitive tasks, bottlenecks, and handoffs?
Then build the agent around that workflow.
Don't start with:
"What can I make an agent do?"
Start with:
"What problem do I want this agent to solve?"
That's where the good ideas come from.
And here's the interesting part.
If a task can be performed through a computer, there's a good chance an agent can perform at least part of it.
The trick isn't figuring out what agents can do.
The trick is figuring out what they should do.
