// SOFTWARE
Where automation ends and autonomy begins
6 min readnijitech
Rule-based automation works as long as the input arrives the same way every time. The moment it does not, it stops. Where the boundary sits and which approach belongs where.
Full post
The word automation is used for two different things, and the difference is large enough to decide whether a project succeeds. The first is rule-based automation: "if this, do that". The second is autonomous flow: a system given a goal that works out the steps itself. They are not rivals; they solve different problems.
Where a rule engine stops
Rule-based automation works flawlessly as long as the input arrives in the same shape every time, and it is both cheaper and faster than an autonomous system. The problem is that the real world does not send the input the same way every time.
Where rule engines typically stop
- Free text — a request a customer wrote in their own words
- The same document arriving with a different layout from every supplier
- A missing field: the rule expects data that is not there
- A situation never seen before
At that point there are two options: write a new rule, or hand the decision to a model. The first is quicker in the short run, but as the rule count grows maintenance becomes impossible — in a system with three hundred rules, nobody knows which rule catches which case.
What autonomy gives and what it takes
An autonomous flow will try to find a way through a situation it has not seen. In return it asks for two things: verification and ownership. Without verification there is no telling where a wrong decision was produced; without ownership there is no telling who reverses it.
An agent and a workflow are not the same thing
Two concepts that get conflated. A workflow is a chain whose steps are known in advance; a model may be used at each step, but the order is fixed. An agent plans its own steps toward a goal, calls tools and changes direction based on intermediate results.
An agent is more flexible and, for exactly that reason, harder to audit. Most enterprise problems are actually solved by a well-defined workflow; agents should be reserved for work where the steps cannot be known in advance.
What it looks like in practice
In a multi-channel inbox, a rule engine queues a message by channel — that part is automation and should stay that way. Drafting a reply requires understanding what the customer asked; that is the autonomous layer’s job. The decision to send is left to a human. With all three working together the system is both fast and auditable.
In short
When deciding
- If input is standard, stay with a rule engine — cheaper and more predictable
- If the rule count is making maintenance hard, the threshold has been crossed
- If you move to autonomy, build verification and decision ownership at the same time
- If the steps are known in advance, you need a workflow, not an agent
From the glossary: AI agent · Human-approved decision (human-in-the-loop)