Benjamin CarlierBenjamin Carlier

Backend, data and AI. Long rides and bikepacking in between.

The tools I actually use

•5 min read
#development#workflow#tools
A generated, nime-style illustration of a hoodie-wearing developer coding late at night in a cozy modern workspace, seated in a Herman Miller Aeron chair facing dual monitors filled with code. The room is softly lit with warm ambient lighting, plants, posters, and tech accessories, while a nighttime city skyline glows through the window, creating a cinematic and focused atmosphere.

There is a difference between the tools you try and the tools you keep. Over the past year or so, I have tried dozens of editors, terminals, CLI utilities, and AI plugins, and most of them did not survive more than a week or two of real use. What follows is the set that did. It's the things I reach for every day without thinking about it, which is probably the best metric for whether a tool actually works.

The editor#

VS Code, still. I have tried alternatives, and I keep coming back to it. Not because it is the best at any single thing, but because the plugin ecosystem means I can shape it into exactly what I need without fighting the tool itself.

The plugins that stayed:

  • Claude Code - I mostly use Claude Code in my terminal but I sometimes use the extension too
  • ESLint - the standard for JavaScript and TypeScript linting, reliable and well-integrated
  • Ruff - fast Python linting and formatting that replaced a whole chain of slower tools
  • GitLens - inline blame and history that saves me from context-switching to the terminal for git archaeology
  • Night Owl - I put this theme on everything, and I mean everything, every editor, every terminal, every tool that lets me change the color scheme

The terminal#

I switched to Ghostty a few months ago and never looked back. It is fast, it is minimal, and easy to config.

CLI tools#

The default Unix utilities are fine, but there are replacements that are just better in ways that compound over a full working day:

  • ripgrep (rg) - faster than grep, better defaults, respects .gitignore out of the box
  • fd - a better find, simpler syntax, faster
  • zoxide - a smarter cd that learns your most-used directories. One of those tools that feels invisible until you try to work without it
  • jq - indispensable for anything involving JSON, which in my line of work is most things
  • htop - process monitoring that actually makes sense visually, for when I have my whole stack running and need to figure out which process is making my laptop fan spin up
  • mole - deep clean and optimize your Mac from the terminal, replacing a handful of paid apps I never wanted to install in the first place
  • glab - GitLab CLI, because work uses GitLab. I use GitHub for my personal projects though
  • rtk - filters and compresses command outputs before they reach your LLM context, saving tokens with almost no overhead

Data tools#

DataGrip with the AWS Toolkit and Big Data Tools plugins handles everything database-related. I have tried lighter alternatives, but for the kind of work I do. For querying across multiple databases, working with large datasets, managing connections to various AWS services, nothing else comes close

Claude Code plugins#

  • Linear - because we use Linear at my company
  • Superpowers - brainstorming, planning, and TDD flows with minimal overhead
  • Context7 - real-time documentation access that keeps me from copy-pasting docs into context manually
  • Serena - codebase indexing that makes Claude Code actually useful on large projects
  • Caveman - compresses agent output by roughly 75%, making responses faster and cheaper without losing technical accuracy
  • Colgrep - semantic code search that combines regex filtering with AI-powered ranking, right from the terminal

With Caveman, Colgrep, Serena, and rtk combined, I never find myself reaching Anthropic's Max $100/month plan limit. I would strongly recommend these tools to anyone looking to get more out of their token budget.

The small things#

Boop for quick text transformations, Base64 encoding, JSON formatting, URL decoding, all the little conversions you need five times a day but do not want to write a script for.

OrbStack as a Docker Desktop replacement. Lighter, faster, less intrusive.

Whispering with Nvidia's Parakeet multilingual model for voice input when working with Claude Code. I'm still not fully comfortable talking to my terminal, but when I do, it works surprisingly well.

What I tried and moved on from#

I experimented with VibeKanban and Paperclip for managing multiple AI agent workflows, and they are well-designed tools for what they do. But I found that I prefer running multiple Claude Code windows side by side and jumping between them manually. It is less structured, but the directness of it suits how I think and I can see everything and intervene immediately.

I have been using this multi-window approach extensively for mypy migration, dependency updates, and setting up Renovate on more than 10 microservices, and it has held up well enough that I stopped looking for something more sophisticated.

The pattern#

The common thread: fast, focused, and out of the way. The tools that survived are not the most feature-rich or the most impressive, they are the ones that removed a specific friction point without introducing new ones.

If you want to keep one phrase out of this: the best tool is the one you stop noticing.

Previous PostEscaping the screen