Rules or Reasoning? How to Choose Between Automation and AI Agents
Why over-engineering with AI can hurt more than it helps, and how to build smarter workflows by knowing when to trust rules and when to invite intelligence.
Everywhere you look, someone’s talking about AI agents.
From YouTube demos to LinkedIn posts, it feels like every automation tool has suddenly rebranded itself as “agentic”, and you can find tonnes of articles on Substack or Medium with titles like “How I build an AI Agentic Workflow in 7 days and sell it for $10,000”.
But amid the buzz, something important is getting lost: most people don’t actually need an AI agent. What they need is good old-fashioned automation — reliable, rule-based, and fast.
In this post, I want to unpack the difference, share a real example from my 3D printing company in Hong Kong Beets, and explain why sometimes a simple script still beats a shiny agent.
Automation vs. AI Agent — Same Goal, Different Mindset
Both automation and AI agents aim to save human effort.
The difference lies in how they do it.
Automation runs on rules and structure.
You define the triggers and steps, and the system executes them every time with clockwork reliability like:
If X happens, do Y.
Developers or Vibe coders can build automation easily in Python, Javascript, Google App Script, etc. Tools like Zapier, n8n, etc are great for as no (or little) code approach. All these solutions are predictable, efficient, and easy to debug.
AI Agents, on the other hand, operate on goals and reasoning.
You tell them what to achieve, and they figure out how to do it — sometimes in creative, sometimes unpredictable ways.
They can read context, interpret meaning, and adapt dynamically.
Think AutoGPT, ChatGPT AgentKit, or AI copilots that plan and act autonomously.
In short:
Automation is rule-driven — you tell it exactly what to do.
AI Agents are goal-driven — you tell it what to achieve.
Both have their place. The trick is knowing which mindset your problem requires.
Choosing the Right Tool — The Simple Rule
If your workflow or output is predictable — built on clear rules, steps, and structured inputs — traditional automation will almost always serve you better.
Here’s a quick comparison:
Data entry, notifications, task routing
→ Best with Automation — Fast, reliable, easy to maintain.
Text interpretation, classification, summarisation
→ Use AI — Handles ambiguity.
The rule is simple. However, our world is not simple and so in most cases, you will encounter both scenarios in a single problem. Guess what? You don’t actually need to choose between both! See below for a simple example.
Real Example from Beets – Email Automation in Action
At Beets, our 3D printing company, we receive all kinds of emails through a shared inbox — from online order notifications to general customer enquiries.
To manage them efficiently, we built an automation that scans each incoming email and performs different actions based on its content.
The goal: Automatically create Jira tasks and assign to the right teammates for messages that need follow-up, and ignore those that don’t.
We quickly realised there were two very different kinds of emails:
1. Structured and predictable
These are things like online order notifications — fixed subject lines, consistent content, and clear patterns.
For these, a simple regex check in Google Apps Script was enough.
Once matched, the automation instantly created a Jira task and assigned it to the right team.
Fast, accurate, and 100% deterministic.
2. Unstructured and ambiguous
Then there were the messy ones — customer enquiries, sales pitches, newsletters, and random marketing emails.
There was no fixed format. Some required action; some didn’t.
Trying to catch them all with rules would have been impossible.
That’s where we introduced AI selectively.
Instead of using an AI agent for the entire flow, we used automation as the backbone.
When the email didn’t match any known rule, we passed it to an LLM for classification — to decide whether it was a real customer enquiry or just noise.
If it was relevant, the automation created a task; if not, the automation will do nothing.
Sometimes, the LLM would make mistake - skipping a real customer enquiry or creating a Jira task for a cold sales pitch. Yet it’s over 90% correct and when it isn’t, it’s not the end of the world as we will eventually pick up the customer email and we can just delete the cold sales pitch Jira task.
The result?
A hybrid system that was both reliable and intelligent — automation where rules worked, AI where judgment was needed.
That’s the essence of how AI and automation can complement each other.
Why Simplicity Still Wins
There’s a temptation to “AI-everything” these days.
But the more autonomy you add, the harder things become to debug, secure, and maintain.
Traditional automation remains powerful because it’s:
Faster to build — no prompt tuning or model setup
Cheaper to run — no costly tokens or API call
Stable and easier to trust — transparent, predictable and explainable
When something goes wrong in a rule-based system, you can trace it.
When something goes wrong in an AI agent, you may spend hours wondering why it decided to do that.
Sometimes the smartest thing you can do is keep it simple.
Closing Reflection — The Triple-A Way
Let’s close off by bringing everything back to a “Triple-A” context.
Agility isn’t about using the newest tool — it’s about using the right one for the right task.
Automation gives you control. It’s the spine — predictable, structured, and robust.
AI gives you adaptability. It’s the brain — reasoning, interpreting, and adapting where needed.
So before building your next AI agent, ask yourself:
“Do I really need that — or I just need a script?”
Obviously a script does not sound that sexy, but it just works.