My thoughts on AI Agents

Should you ride this wave?

Is it the time for most people to tinkle around with AI Agents? Is it capable of solving real-world problems?

Table of Contents

My experiments

I’ve been using Windsurf Pro for the past couple of weeks, and my feelings on it vary from task to task.

  • ✅ It shines when you are working on a technology that you don’t fully grasp, and you need help to understand it (more on this later).
  • ✅ It shines when you need to perform some mechanical work (like formatting code, refactoring, creating multiple files, adding tests, etc).
  • ✅ It shines when you already have something similar to what you want and you tell it to base the solution on that.
  • ✅ It shines when you think about the solution and provide a lot of context.
  • ❌ It doesn't shine when your prompt is vague.
  • ❌ It doesn't shine when you work on a complex project.
  • ❌ It doesn't shine when you don’t think about what you want it to do.

Is it helping me right now?

It does help me, but not always. I believe it’s currently on the stage where you need to be very mindful about how much time will you allow it to help you with a task, but you need to use a stop-loss strategy if you see it’s not going in the right direction.

I do want to do more experiments with it. In particular, I want to wrtie a really complete set of Contributing Guidelines, and check how much it can follow them without derailing.

The vibe coding era

This is something that has been both amazing, and perturbing in my mind for the past few months. I am seeing teenagers on X, bragging about vibe coding, and shipping 10x faster than really accomplished engineers. This is something that is really encouraging, as it is elevating the product competition. However, I do worry that people (and particularly the current Junior Engineers) will start to lose focus on learning the fundamentals for coding.

My Advice

I think the best way to approach this is to:

  • Use AI Agents when you need help with a task, but be mindful of the time you allow it to help you.
  • Here is the fun part, you don’t want to mindlesly accept whatever the Agent spits out. You want to mindfully ask questions about the solutions, understand what it’s written, and discuss with it. You need to be using this as a learning tool, not as a computer-slave that just follows instructions.

We are still pretty early into this new AI era, I expect to be new products, new paradigms. 3 years ago, we didn’t have AI Chats, last year, we didn’t have agents, so I don’t expect this to be the latest or predominant form of AI for long time.

Paraphrasing what Cal Newport mentioned in a recent podcast:

Whenever a revolutionary AI product appears, people will start to use it, you will hear about it in the office, it will take you 5 minutes to learn, and then you will be using it all day long. (Think of the time when the email appear).

So, if you want to wait for that time, you will be ok, if you want to geek-out and use the products of the process, it is also ok!


Related Articles


My thoughts on AI Agents | manu.show
Tags: thoughts
Share: X (Twitter) LinkedIn