What I Built on Day One So My Client Wouldn’t Need a Rewrite

One way I save my clients tons of time is by spotting which business rules are likely to change and building for that from day one. I learned to look for those rules by working across very different products and seeing how often a requirement changes once people actually use the software.
On a product I’m building for a client, several features use AI under the hood. The client was very clear about how the workflow should operate: a person had to review and approve the result of every phase before the next one could start.
The client was right to ask for approvals
At the beginning, they hadn’t seen what each phase would produce. Reviewing the results gave them a chance to catch mistakes, understand what good output looked like, and decide where they were comfortable relying on it.
I also suspected that the same people who wanted to check every phase at launch might ask for something different once they had seen enough results. They might be comfortable letting the first two phases run automatically while still wanting someone to sign off on the final result. Over time, they might want the whole workflow to run on its own.
I made the approval rule easy to change
I built the workflow so each phase can advance automatically, wait for one approval, or require multiple approvals. We could start with the manual checks the client asked for and change them phase by phase later.
The part that produces a result doesn’t have to change just because the client decides they no longer need to approve it every time. That’s the decision I wanted to keep open. I didn’t need to guess which phases they’d automate or when they’d be ready.
Four months later, they asked for automation
They wanted to start automating some phases, with the goal of eventually automating the whole workflow. I didn’t have to go back to them with a rewrite. I told them we should keep testing manually and, once we trust the output of a phase, turning on automation for that phase is a one-time toggle.
That puts our time where it belongs: checking whether the results are good enough to automate, rather than rebuilding the workflow just to find out.

