Skip to content

Month: February 2025

LLMs to spot design flaws

I just discovered another interesting use case for current LLM tools. I use Claude.ai at the moment it’s the best in class. I use the free version.

When you design on a document, let’s say a GDD or a one pager for a feature, you should cover the most possible edge cases. In this way, your coders will have everything pretty much clear by reading the document.

If you submit a document to Claude, then you can ask it to simulate the coding in Unity or Unreal Engine for that GDD. It will spit out pseudo code and reasonings. There will probably be lots of hallucinations and things unuseful. Still, there are good chances to discover something you didn’t considered.

Working in a team is self-discovery

Today I discovered something more about myself, thanks to the creative director of the project I am working on with a company.

I tend to not insist too much on my vision. I explain it, defend it, and usually that’s it. When I see too much resistance from the other side, if I have no real power over the decision, I desist and try to meet the boss’ vision.

This is good, but it can lead to a passive-aggressive way of communicating. “I will do like you say, but I do not agree”. At first it may seem like there’s nothing wrong with that, butthe issue is that:

  • it looks like a “ok, whatever” and can damage the relationship
  • it is vague, proposes no real solution, and can damage the project
  • It is not informative enough for the team to make choices on that

Today I have learn something more about myself. Something I want to fix. And that’s why I prefer to work within a team.