- A simple automation follows fixed rules: when this happens, do that.
- An AI agent reads, judges and writes, so it can handle messy inputs like emails and documents.
- Most good systems combine both, with people approving anything sensitive.
Sarah runs operations at a wholesale distributor that supplies around three hundred independent shops. Two jobs eat most of her team's mornings. The first is copying new orders from the web shop into the stock system and emailing the warehouse a picking list. The second is working through the shared inbox, where shop owners send everything from delivery queries and complaints to orders typed into the body of an email, sometimes with a photo of a handwritten list attached.
Both jobs are repetitive. Both could be automated. But they need very different kinds of automation, and understanding why will save you money and frustration when you decide what to build.
What a simple automation is
A simple automation follows rules. When something happens, it does a fixed set of things. People often describe it as "if this, then that".
- When a web order comes in, create it in the stock system and send the picking list to the warehouse.
- When an invoice is seven days overdue, send the first reminder.
- When a form is submitted, add the contact to the CRM and notify the sales team.
- Every Monday at 8am, pull last week's sales into a report and email it to the directors.
These automations are reliable, fast and cheap to run. They do exactly the same thing every time, which is what you want for a task that should never vary. They are also easy to check, because you can look at the rules and see what will happen.
Their limit is that they cannot cope with anything they were not told about. If the information arrives in a different format, is missing a field or needs interpreting, a simple automation will either stop or do the wrong thing. It does not understand what it is handling. It only moves it.
Sarah's first job is a perfect fit. Web orders arrive in a consistent structure, the steps are always the same, and there is no judgement involved.
What an AI agent is
An AI agent can do three things a simple automation cannot: read, judge and write.
- Read: it can take in messy, unstructured information like an email, a PDF, a photo of a document or a chat message, and work out what it says.
- Judge: it can decide what kind of thing it is looking at and what should happen next, within limits you set. Is this an order, a complaint or a question? Is it urgent? Does it need a person?
- Write: it can produce a sensible reply, a summary or a structured record, in your tone and using your information.
Sarah's second job needs an agent. An email that says "same as last week but double the oat milk and drop the crisps" requires reading, understanding context and checking the previous order. An agent can draft that order, reply to the shop to confirm it, and send anything that needs a human, like a complaint about a damaged delivery, to the right person with a summary already written.
What an agent should not do on its own
Agents are capable, but they are not infallible, and they should not be given free rein. A well designed agent works within clear boundaries.
- It handles the routine cases it has been set up for.
- It hands anything uncertain, sensitive or high value to a person.
- It records what it did and why, so someone can check.
- It uses only the systems and information it has been given access to.
That is what keeping a human in the loop means in practice. The agent does the reading and the first draft. A person makes the calls that carry real risk, like issuing a refund, agreeing a price or responding to an angry customer.
How to tell which one you need
A few questions will usually settle it.
- Does the input always arrive in the same format? If yes, a simple automation is probably enough. If it arrives as free text, attachments or varied documents, you likely need an agent.
- Is there one right answer every time? Fixed rules suit fixed answers. If the right response depends on what the message says, that is a job for an agent.
- Does the task involve writing something new? Moving data between systems is automation territory. Drafting replies, summaries or tailored messages is agent territory.
- How bad is a mistake? For high stakes steps, use simple, predictable rules where you can, and make sure a person approves anything the agent is unsure about.
Examples across a business
- Sales: routing a web form to the right salesperson is a simple automation. Reading the enquiry and writing a tailored first reply is a job for an agent. Our sales page covers both.
- Finance: sending reminders on a schedule is a simple automation. Reading a client's reply and pausing the reminders because they have promised payment on Friday needs an agent. There is more on this in how to get paid faster.
- Customer support: tagging tickets by a form field is automation. Understanding what a customer is asking and drafting an answer from your help content is an agent. See our customer support page.
- Hiring: moving applicants between stages is automation. Reading CVs against a role's must haves and summarising each candidate for the hiring manager is an agent.
Most good systems use both
In practice, the best setups combine the two. The agent handles the messy front end: reading the email, understanding it and deciding what it is. Simple automations handle the reliable back end: creating the record, updating the stock system, sending the confirmation and logging everything.
For Sarah, that means one joined up system. Web orders flow straight through on rules. Email orders are read by an agent, turned into the same structured order, checked against the shop's history and passed into the same rules. Anything odd lands with her team, already summarised. Her team stops retyping orders and spends their mornings on the customers who actually need a conversation, and fewer orders go out wrong because nobody is copying them by hand.
Using both also keeps cost and risk sensible. There is no reason to use an agent for something a simple rule does perfectly well, and no reason to force rigid rules onto a task that needs judgement.
Talk to us about what fits
If you are not sure whether a task in your business needs an agent, a simple automation or a mix of the two, we can help you work it out. The Orqen team builds both, inside the tools you already use, and always with your people in charge of the decisions that matter. Get in touch with us and tell us what your team repeats.