AI agent vs automation
Kahu team·Updated 18 Sep 2026·5 min read
Automation follows fixed rules you write in advance: when X happens, do Y. An AI agent is given a goal and works out the steps itself, adapting when something unexpected comes up. Use automation for tasks that are identical every time; use an agent for tasks that are similar but never quite the same.
Automation: if this, then that
Business automation has been around for decades. A form is submitted, an email goes out. An invoice is 7 days overdue, a reminder is sent. Tools like Zapier, Make and the built-in rules in most software do this well. The strength is predictability: the same input always produces the same output. The weakness is the same thing. The moment a customer writes "actually can I change my order?" the rule has no idea what to do.
AI agents: here is the goal, work it out
An agent starts from an outcome. "Make sure every enquiry from last night gets a helpful reply by 9am." It reads each message, decides what a good reply looks like, checks your prices or availability if needed, drafts, and sends or asks. Two enquiries never get identical treatment because two enquiries are never identical.
| Rule-based automation | AI agent | |
|---|---|---|
| You provide | Exact rules | A goal and context |
| Handles variation | No | Yes |
| Predictability | Total | High, not total |
| Setup effort | Per rule | Once, then it generalizes |
| Best for | Invoices, reminders, data sync | Enquiries, follow-ups, content, scheduling |
| Failure mode | Silently does nothing | Does something slightly off, so review matters |
Which tasks belong where
- 1Identical every time → automationSending a receipt. Moving a booking into the calendar. Reminding someone 24 hours before. Write the rule once and forget it.
- 2Similar but varied → agentReplying to an enquiry. Following up a quote. Writing this week's posts from this week's photos. Each one needs a little judgment.
- 3High stakes → agent with approvalRefunds, price changes, anything published under your name. Let the agent draft, you approve.
They work best together
The most useful setups use both. Automation does the plumbing: capturing the enquiry, logging it, triggering the agent. The agent does the thinking: reading it, replying, deciding whether a follow-up is needed. Then automation takes over again to log the outcome. Vendors who sell "AI automation" are usually describing this combination.
Where Kahu fits
Two Thursday gaps at 11:00 and 14:30. Drafted messages to 6 waitlisted customers.
Updated the hours on your site and Google profile to Sat–Sun 8–2.
One post about the new hours, using your Sunday counter photo.Frequently asked questions
Is Zapier an AI agent?
Zapier is an automation platform. It has added AI steps that can draft text or classify messages, which brings it partway toward agent behavior, but the flow is still a fixed rule you design.
Which is cheaper?
Automation is cheaper per task. Agents cost more per task but replace tasks you would otherwise do by hand, so the comparison is against your own hours.
Can an agent replace all my automations?
It could, but you would not want it to. Use fixed rules wherever a fixed rule works; save the agent for the parts that need judgment.