When not to use an AI agent: five situations
Agents are powerful, but they do not suit everything. Five situations where you are better off putting a person on the job.
A company wants an agent to rewrite its entire customer portal in one go. A week later there is a proposal of thousands of lines. Nobody can review it. Nobody dares to approve it. The week is lost.
A good adviser also tells you when something won't work. These are the situations in which we advise against using an agent.
1. When nobody knows what the result should be
An agent executes. It does not decide what your product should do. If your team doesn't know that yet, start with a conversation, not with an agent.
2. For major architecture decisions
Which database, how systems talk to each other, how your application grows over the next five years. That takes overview, experience and knowledge of your business. An agent can help with research and a first version, but the decision is human work.
3. During an active security incident
If something is going wrong right now, you need a specialist who works fast and with full attention. After the incident, an agent can help roll out the fix everywhere.
4. When review takes longer than the work itself
Some tasks are so small or so specific that you do them faster yourself than you can write a task and review the result. Then you gain nothing.
5. When there is no way at all to check that it works
No tests, no test environment, no way to try the result. Then the first step is to build that check. Only then the agent.
What that means for your scan
In our free scan, we look at where agents save you time, and where they don't. Sometimes the honest answer is that there is little to gain. Then we say so.