AI as a mentor, the Socrates way

Someone taking handwritten study notes next to an open laptop, a quiet setting for using AI as a mentor
After almost 20 years in IT, I’ve done a lot of training. As a Software Engineer, I need to stay on top of new technologies. That means following training and learning paths. However, it is not always easy to keep up with that pressure. I recently earned the Claude Certified Developer – Foundations. I’m grateful to Zartis for making it possible: they funded it, gave us access to a learning platform and encouraged us to take it. This certificate is a rewarding achievement for me. Preparing for it (and some other training) is where I tested a different way of learning.What are the key points of this article: benefits of using a skill in Claude, how to use them as a personal mentor and don’t trust everything AI tells you, check it.Have you started a training course you couldn’t finish? Have you followed a podcast or started an online bootcamp on cutting-edge technologies only to drop it after the first week? Have you clicked on a YouTube tutorial with a lot of energy at the beginning and then abandoned it after thirty minutes of watching someone else’s mouse move around their desktop? I know how it feels because I’ve been there too. It all starts with good intentions, but soon it becomes boring. Eventually, it ends up as one more item on the long to-do list of learning stuff. I couldn’t let that happen with the Claude certification.

There are many ways of learning tech nowadays, and most of them share one trait: they’re passive. You watch, you read, you listen. You can follow along and build a project yourself, and that’s a step up, not passive anymore. But if nobody sees what you’ve built, you’re still on your own. Nobody checks whether you’re making progress. The challenge is to stay focused. It is easy to postpone it when you have a project deadline, a report for your manager or a demo to prepare. It is easy to disengage from the learning process.

 

Mentorship and Socrates

At some point in this vicious circle, I thought back to my school days. A good teacher wouldn’t have let me disengage. What made them good? They were people who would guide you and show you new things, but also challenge you when you went down the wrong path. That reminded me of Socrates, the renowned philosopher from Athens. He would take a walk or sit with his pupils, ask them questions, test them and push them to think for themselves. The method is 2,400 years old, and it’s still used in schools and universities today. It may be one of the oldest forms of mentoring.

But Socrates isn’t on call and you can’t always find a real mentor. You don’t necessarily have someone at hand to talk with you and ask the right questions at the right time. Over my career, I’ve had a few mentors, and it has been incredibly rewarding learning with someone I could ask, and who made me question my own thinking.

 

How I built the skill

So I decided to recreate that dynamic as a skill in Claude. I won’t explain here in detail what a skill is. You can refer to Anthropic documentation about them. However, I can tell you how I built mine. I wanted to feel supported in my training. The way I’d usually start a conversation would open with a sentence like this: “I’m following a course on (some topic), my goal is to (get a certification/learn something)”. So you can add something like this to the description: “Use when the user wants to learn, study or prepare for a training, course or certification exam (e.g. ‘follow a course on X’, ‘learn about X’, ‘get a certificate on X’)”. This defines when the skill will activate.

Then I would like to have at least these three conditions for that coaching:

  1. “Don’t give me the answer straight away, unless I ask for it”
  2. “Ask questions and challenge my knowledge”
  3. “Look for grounded sources of information (such as official documentation or course content) to confirm before arguing”

These conditions go into the skill’s instructions. They will define what happens when the skill is activated. You can make them more detailed, or write your own, but this is enough to show how it works.

These rules allow me to follow a natural learning process. Knowing the answer upfront does not help me to learn a new concept. It is more useful for me to have a discussion starting from what I already know with “someone” who knows the topic, and drives the conversation so I can reach the conclusion myself. When the idea is already in my head, I can reinforce it by explaining it in my own words and getting corrections from Claude. One very important condition to notice here is the third point. I ask Claude explicitly to always base the conversation on real sources. That way I make it harder for Claude to assume things without the right context, raising the chances of catching an AI hallucination.

 

How I used it

Preparing for the Claude certification is where I used this approach the most. The conversations were not always soft and comfortable. Claude pushed back on many of my answers, and I pushed back too. I asked for more explanations and, more than once, for the documentation behind what it was saying. Sometimes I was just wrong, and the conversation made it clear why. Other times I had to look for more information myself before I could answer. And once, Claude made an assumption I didn’t agree with. When we checked the documentation, I was the one who was right. I had to push my own limits a few times, but that arguing was the most useful part of the whole process. It also helped me notice a pattern in the exam questions: some of them had two or more answers that looked valid, but one word in the question made one of them the wrong choice. When I took the exam, I was already looking for those words.

I’ve used it in other trainings too, and I can share some examples. There was a training I followed on AWS. The contents were not very difficult, but because I’m an engineer I focused on making it work, not the cost. In the context of this training, keeping the costs under control was one of the key topics. There were a few times that Claude pushed back on my answers because of this skill. It kept on reminding me to be aware of the economic implications of every choice. It kept me focused on costs.

Later on, I decided to take a Delivery Manager training. This is quite a different knowledge area from mine. Claude also helped me build a glossary to understand a list of terms about projects, costs and business conversations. But the most meaningful contribution from it was to help me switch my mindset: from an engineer to a manager. There was a scenario about a project deviating substantially from the original estimate. My initial answer reflected my background as an engineer, jumping into the project and helping it move forward. Claude acknowledged my engineer’s background, and gently asked me to rethink it as a proper Delivery Manager. The right answer in this context was to gather metrics to confirm the deviation and then flag it to the stakeholders and the Account Manager. Doing a good job here was not only about the knowledge you need, but also about the goals to set, the right attitude and the proper approach to make it successful. It was extremely helpful to discuss the whys and the hows on top of the concepts.

 

Refining the skill

In general, I got a good feeling of being supported during the process. It is good to be aware that this is a tool and it performs as well as the people using it. It can’t fully replace a real mentor, but it can help when there’s no one available. The skill can also be refined over time. You should add more conditions to the initial ones. Mine started as a few lines, evolved into a fuller version with modes and progress tracking. That is also a learning process.


So, what will you do in your next training? Are you reading, watching or listening as a passive receiver? Or will you try having an argument with Claude about the best practices or the mindset you’re trying to learn? Build your own skill and share your results in the comments.

 

Author

Damián Álvarez, software engineer at Zartis and author of this article on using AI as a mentor Damián Álvarez is a software engineer specialising in quality backend code and comprehensive documentation. He works mainly in C#/.NET and SQL, with experience in Angular, React, and two years in platform engineering across AWS, Terraform, Ansible, and Python. A Claude Certified Developer – Foundations, he uses AI-assisted development with Claude to speed up how he builds and delivers software, and is exploring what these tools mean for our profession and society.

Newsletter

Zartis AI Review

Your monthly source for AI and software related news.