Deleted a few dozens Shortcuts today. All sort of Shortcuts: simple ones, utility ones, more complex ones. Many of them aren’t needed anymore because of my AI usage. I’m curious to see how Apple will make the Shortcuts landscape evolve in the age of AI.

Sometimes I test a local LLM on my M2 MacBook Air, which isn’t very powerful for this task. After a few prompts, it gets quite hot.🌡️ A MacBook Pro, with fans, would handle it better. Now, think about millions of people prompting LLMs in data centers worldwide, with much more powerful models responding and consuming huge amounts of energy. No wonder some parts of the world are experiencing an energy crisis because of AI.

This morning, I realized that having a self-hosted version of n8n comes with some notable limitations. For example, with the SaaS version, you can link your n8n instance to a GitHub repo so you can use version control right in GitHub and manage issues and all that. More details from Claude AI:

n8n has a native Source Control feature (available in Enterprise / Self-hosted Pro plans) that connects your n8n instance directly to a Git repo: Push/pull workflows, credentials stubs, and variables to/from a branch Supports environment promotion (e.g. dev → staging → prod branches)

If Google’s Gemini will be the core intelligence behind the “new Siri” this fall, as per Google’s recent statement, should I begin exploring it to understand its personality? 🤔

I believe I decided to build my web apps at the right time1, because if I did it now, it would take two or three times longer due to the current credits consumption rate enforced by Anthropic. I made a little tweak to my dashboard web app this morning, and I’m already at 36% for the current session. It’s really that bad. I wouldn’t pay 200$ a month to get more credits.

  1. In the first three months of 2026. ↩︎

This morning, I realized that managing open issues and bugs on GitHub is advantageous: the more detailed each open issue is, the more effective it serves as a prompt when importing into Claude Code to initiate a new bug-fixing session. Claude Code can also close the issue at my request and link it to a specific GitHub commit.

I’ve been testing the latest release of the Claude Desktop app, and I must say, more than ever, I prefer integration to splitting features across many different apps. I’m also leaving the CLI behind for now.

For all my pending issues across my different Claude Code and web apps, I just realized I could (and probably should) use GitHub for issue tracking instead of Craft. Additionally, I could develop a web app to monitor and manage all open issues across my repositories!

Today, I tested Ollama and Locally AI more extensively on my M2 MacBook Air, and it was quite demanding. It’s no surprise that a serious local AI setup requires an Apple M5 Pro or M5 Pro Max with at least 16 or 24 GB of RAM. My M4 Pro Mac mini has 24 GB, and I could use it remotely through an SSH session. This experimentation puts any plans to replace my aging M2 MacBook Air into perspective.