AI tools make building software faster than ever. As a software developer of 20 years, that’s part of what worries me. You can build something quickly and still end up with an expensive, rigid system you have to maintain.
Before building another system, I’d start with one painful workflow inside the business.
Look for work people hate doing, something that takes forever, or a task that keeps falling through the cracks. That gives you a specific problem to work on. The next question is how an agent could help carry out that work.
Describe the work before building the system
The approach I’m interested in is agentic workflows: using an AI agent to understand a workflow and do the heavy lifting. English becomes part of how you describe and adjust the work, rather than building the whole system in a programming language such as C# or PHP.
That’s also what appeals to me when something needs changing. In this approach, you can explain what isn’t working to the AI and ask for an adjustment. The possibility of changing the workflow through conversation is a reason I’d explore it before committing to another rigid system.
Start with the task your team keeps avoiding, the slow process, or the work that gets missed. Describe that one workflow clearly, then figure out where an agent could help.