How I use LLMs and other generative AI tools

Claude Code and Codex

I use AI a lot at work, both for software development and for work projects. But that is a separate story.

In this post, I want to share how I use AI for my personal blog and personal projects.

AI for blog and YouTube visuals

I use AI to create images for my blog posts and thumbnails for my son’s YouTube channel, RostyLife.

Most of the time, I use AI as an image editor rather than generating images from scratch. I usually provide real photos as references, describe the result I want, and ask the model to help with composition, background, lighting, or the overall look.

I recently used this approach for a post about Kizzikids Cardboard Tools.

Sometimes I generate an image from scratch when I need an illustration for a specific topic and do not have a suitable photo.

AI as an English writing assistant

English is not my native language, so I also use GPT models to review my blog posts.

I do not ask AI to write posts for me. I write them myself in English and then ask ChatGPT to check the grammar, suggest corrections, and explain my mistakes.

This helps me publish clearer posts while also improving my English over time.

The ideas, stories, opinions, photos, and experiences in my posts are still mine. AI helps me express them better.

AI for personal software projects

I also use AI when I work on personal software projects.

I use both Claude Code and Codex. For personal projects, I prefer working directly in Claude Code instead of building or using a separate agent harness.

My workflow usually looks like this:

  • I use Opus to think through the architecture, create an implementation plan, and make important technical decisions.
  • I use Sonnet for implementation.
  • Codex reviews the code.
  • Opus looks at the review and decides whether the suggested changes are worth making.
  • If they are, Opus gives Sonnet another task.

I created a Claude Code skill that can repeat this process for three to five iterations when a task needs more than one implementation and review cycle.

This workflow makes personal projects much faster to build. It also makes it easier to test an idea quickly and see whether it is worth developing further.

In the past, trying the same idea could take much more time.

To let Claude Code use Codex as a subagent, I use the OpenAI Codex plugin for Claude Code.

What AI models and tools do you use, and what do you use them for?

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *