In 30 seconds
For any step in a workflow, prefer the simplest mechanism that solves it reliably. Manual, rule-based, AI-assisted and agentic are all valid outcomes.
The most advanced option is rarely the best one. The most appropriate option usually is.
Why this comes up
It is easy to reach for the most capable tool because it is the most interesting. An agent that reads, decides and acts feels more impressive than a two-line rule.
But complexity has a running cost. Every added capability is something else that can fail quietly, something else to test, and something else to fix when a system or a policy changes six months from now.
The principle in one line
Notice the last part. A birthday reminder and a payment approval need very different reliability. The principle is not "always do the cheapest thing." It is "match the mechanism to the real requirement, and no more."
A small example
A team wants to flag urgent support emails.
- The simplest version: flag anything containing "urgent" or "down". Crude, but catches most cases in an afternoon.
- A better version: an AI reads tone and content and estimates urgency.
If the crude rule catches 80 percent and a person reviews the rest, that may be enough for now. The AI version is worth building when the missed 20 percent is actually costing something. Start simple, then add capability where it pays.
What this does not solve
Choosing a simple mechanism does not excuse skipping reliability work. A simple rule still needs testing, ownership and a fallback. Simple does not mean careless.
Good enough to move on when
- You considered a simpler option for each AI step.
- You can explain why the added complexity is justified.
- The reliability of each step matches what the task requires.
