Lately, there’s been a lot of buzz around agents, and for good reason. Tools like Cursor, ChatGPT, and Perplexity have completely changed the way I work. They’re groundbreaking, no doubt about it, but it’s easy to assume they’re all-seeing and all-knowing. The truth is, agents are just one of many tools in our toolbox. And sometimes they’re not the right tool for the job.
What They’re Good At
At their core, agents use transformer technology to really understand dynamic and unstructured data. This makes them incredibly good at:
- Generating dynamic content: Just feed them a prompt, and they can produce creative and varied outputs.
- Data transformation: Whether it’s converting English to Chinese, Ruby code to Python, or even transforming messy data into neat tables, agents handle it well.
It’s interesting to think about how much of our day-to-day work involves these same processes. Software engineers, for example, translate product requirements into code. Doctors turn patient conversations into clinical notes. Marketers adjust strategies in real time based on shifting objectives. In these cases, agents can be a massive time-saver.
What They’re Not Good At
However, agents have their limits.
- They can only pattern match. They can only work with what they’ve seen before. This means they’re not great at coming up with entirely new workflows or strategies—they’re more about pattern matching and emulation. Human intelligence is still needed to come up with new strategies.
- They are non-deterministic. If you have sensitive, high-risk tasks that require strict rules (like bank transaction processes), relying solely on an agent isn’t the best idea. In these situations, traditional automation that follows set rules is a safer bet.
How to Think of Agents
I like to think of an agent as a regular person who has instant access to all the resources on the web. If you wouldn’t expect a normal person to solve a problem in a reasonable amount of time with internet access, then don’t expect an agent to either. They’re not geniuses—they’re helpers.
That said, if the task is something that’s doable by you but takes a lot of time (especially if it’s repetitive), agents can be a perfect fit. Just be prepared for a few small errors along the way. And if you’re dealing with high-risk processes, it might be best to use agents to draft automation (like writing code) that you can then review and test before putting it into production.
In Conclusion
Agents are groundbreaking because they give us new ways to handle unstructured data and tackle translation tasks. But remember, they’re just one tool in your toolbox. Sometimes they’re the right tool for the job, and other times, you might be better off reaching for something else.
