Agents & Automation
Not everything needs an AI agent.
Agents are useful when the work requires judgment and multiple steps. Adding one to a simple process can make a good workflow unnecessarily complicated.
AI agents are useful.
They are also becoming one of the most overused answers in AI conversations.
A workflow does not become better just because an agent sits in the middle of it.
Start with the decision the system needs to make
Imagine a new enquiry arrives.
If the next step is always: create a record → assign it → send a confirmation, you probably do not need an agent.
The process already knows what to do.
Now imagine the enquiry needs to be understood first.
Is it sales? Support? Training? A partnership request? Does it need information from another system? Can it be answered automatically or should a person handle it?
That is a more reasonable place to consider an agent.
More autonomy means more design responsibility
Giving a system the ability to decide what happens next can be valuable.
It also creates questions.
- 01What is it allowed to access?
- 02What actions can it take?
- 03When must it stop?
- 04When does a person approve the next step?
- 05What happens when the information is unclear?
Agentic systems need boundaries, not just intelligence.
There is no prize for making every step “agentic.”
Use rules where rules work
Fixed rules are not outdated.
They are predictable.
If a business rule can be expressed clearly and does not require interpretation, deterministic automation is often easier to test and maintain.
Use AI where uncertainty exists.
Use rules where certainty already exists.
The strongest workflow may be mixed
A good system might look like this:
- Automation receives the request.
- AI understands it.
- Rules check permissions.
- An agent prepares an action.
- A person approves a sensitive decision.
- Automation updates the business system.
There is no prize for making every step “agentic.”
Complexity has a cost
Every extra model call, integration, permission and decision path creates something that must be monitored.
The goal is not to remove all complexity.
It is to spend complexity only where it creates value.
Agents are powerful when the work genuinely needs them.
That is exactly why they should not be used everywhere.