AI Recommendations

AI recommendations

Last updated:

There is already more AI advice than anyone can reasonably keep up with, and much of it becomes outdated quickly. This is a practical collection of tools, technologies, people and techniques that I currently find useful.

They are recommendations, not universal rules. What makes sense for you will depend on your work, budget and how much complexity you want to manage.

Models

GPT5.6 Sol (USD100 / month)

My preferred model for 90% of work. It is on par with Claude Fable and Opus 5, but is more token/cost effective.

Claude 5 Opus or Fable (USD20 / month)

For when I want a "second opinion" from a completely different model.

AI Tools

Herdr

I recently switched from cmux to Herdr as my main terminal for running multiple agents across projects. it is similar to cmux in many ways, but I just find it a bit nicer to use. Herdr runs inside Ghostty (or any terminal application), instead of being a fork of Ghostty like cmux.

Pi

My new "harness", that replaces Codex CLI and Codex Desktop as my preferred way to use my Codex subscription. It gives more control over the system prompt and tools, meaning I can configure it the way I like.

"Skill Manager"

A Mac app that I built to track, add, view, edit and compare AI skills across all my projects. The information behind the app is also available to agents via a related skill, making it easy for them to help manage skills. I am still considering whether to release it publicly, but let me know if you want to give it a try.

"Athena"

Another app I built myself. It's a project management tool that is built to be agent-first. Agents working on various projects know when and how to use Athena. They can open, update, classify and close jobs themselves, and link to files and ai session transcripts. They use this tool to track progress and handover work to new agents between sessions. Importantly, I do not need to tell them when to use Athena, they have a skill that guides them.

cmux

My preferred terminal for running multiple agents across projects, with workspaces and notifications that keep parallel work manageable. It's a coding tool, but I use it for ~90% of my AI work.

Visual Studio Code

For reviewing and editing code, and most of my git activities.

Claude Desktop

While I prefer CMUX for most work, the desktop apps provide a cleaner interface for certain tasks, and easier handling of projects and scheduled tasks.

ChatGPT Desktop

Same reasoning as Claude Desktop.

MacOS Tools

Obsidian

Markdown is the preferred language for AI, and Obsidian is a great tool for working with markdown in a "human-friendly" way.

Raycast

A great tool for quickly launching apps, running scripts, and managing workflows, although the new MacOS Spotlight has caught up in many ways.

Aerospace

Working with multiple agents across multiple projects is easier with multiple screens, and Aerospace fills a gap left by MacOS's native window management.

"Jogger"

Another app that I built myself. This one gives me quick access to keyboard shortcuts, CLI commands, scripts and config files that I can now access with a single keystroke instead of having to search all over the place. New apps are added by AI with a simple request via a related skill. (Jogger is short for "memory jogger".)

Web Hosting

Cloudflare

For domain registration, DNS, CDN and security services for Listium, and hosting multiple client projects and side projects.

Render

I switched from Heroku to Render for hosting Listium. It was cheaper, and a much better experience. I use web services, postgres, redis, and cron jobs.

YouTube channels

Nate B. Jones

A daily editorial on major AI news and trends.

Nate Herk

Hands-on tutorials about AI automation, agents and workflows.

Fireship

Short, entertaining updates on developer tools and AI technology.

Theo

Opinionated commentary focused on software development and AI coding tools.