A while ago I noticed that, despite using AI tools every day, I was getting more and more annoyed by them. The models were becoming faster and smarter, but the way I had to use them was still mostly the same: open a chat, explain everything, wait, copy the answer, and go back to whatever I was doing before.
And this is fine when you want to have a long conversation or start something from scratch. But most of the time I did not. I needed to fix one sentence, check one fact, or ask one small question without dropping everything else.
I was tired of turning every small task into a conversation.
I was tired of waiting for a small edit
Let’s say I’m writing an email and notice that one sentence sounds weird. I copy the whole email into ChatGPT and ask it to fix that sentence. It rewrites the whole thing, so now I have to read it again to make sure it did not change my tone or add something I would never say.
I wanted a five-second fix and got another review task instead.
I was tired of waiting for AI to rewrite text that was already fine. I wanted to select the part that needed help, press a shortcut, get a suggestion, and continue writing.
The same applies to longer work. If an answer is useful, it should be possible to put it into a Note or Document and edit it directly. There is no reason to ask the model to generate the whole thing again because one paragraph changed.
I was tired of losing useful answers
Another thing that kept happening was losing useful answers. I knew that I had asked something before, maybe last week, and I knew that the response was good. But I had no idea which chat it was in.
So I would scroll through the history, open a few conversations, search the browser tabs, and usually give up and ask the same question again.
I was tired of searching through old chats. A good answer should not become useless just because I cannot remember the title of the conversation where it appeared.
Lutin keeps conversations, feed activity, Notes and Documents, and memory in the same app. Notes, Documents, and saved memories are searchable. Important things can be kept as actual work instead of being left somewhere in a chat history.
I was tired of switching apps all day
Most questions do not appear while I am looking at an empty chat window. They appear while I am writing code, reading a document, or replying to someone.
The usual flow was to switch to the browser, find ChatGPT, start a chat, paste the context, wait, copy the answer, and switch back. Nothing about this is particularly difficult, but doing it twenty times a day is annoying and it breaks focus every time.
I was tired of switching between apps just to ask a quick question.
Lutin is a native desktop app that opens with a global shortcut over the app I am already using. I can ask what I need and close it when I am done. Opening it does not automatically give it access to whatever is underneath. I still choose what context I want to provide.
I was tired of copying and pasting
Copy and paste is probably the most used AI integration in the world. Text, links, screenshots, code — everything goes through the same little chat box. At some point my clipboard history looked like a log of conversations with AI.
I was tired of copy-pasting context back and forth. Not because copying text is hard, but because software should be able to do better while still letting me decide what I share.
With Lutin, I wanted context to be something I choose and attach when it is relevant. I also wanted actions that need approval to ask first, instead of an assistant silently doing things in the background.
I was tired of the best AI tools being only for developers
There was also another, slightly weird, part of all this. Some of the best AI products were developer tools like Pi and Codex.
They are not just chat boxes. You give them access to a codebase, they read the files, make changes, run commands, check whether the result works, and continue until the task is done. They still make mistakes, but at least they were built to do the work instead of only talking about it.
The problem is that this works especially well for software development. A codebase gives the agent a relatively clear world: the files are structured, the tools are known, and tests can often tell it when something is wrong.
When companies try to bring the same idea to everyone else, products like Cowork and the work features in ChatGPT often feel like developer agents with the coding parts removed. To me, they feel less like tools designed around non-developers and more like attempts to adapt a developer product after it was already built.
Most everyday work does not look like a codebase. It is spread across emails, documents, calendars, notes, screenshots, and conversations. There is usually no test suite that can tell an agent whether it helped. Context, timing, memory, and permission matter much more.
I was tired of the choice being either a chat box or a developer agent adapted for people who do not write code.
What I wanted instead
I did not want Lutin to automate my entire life or replace every app on my computer. I wanted it to remove a few repeated annoyances.
I wanted one shortcut instead of another permanent browser tab. I wanted small edits to stay small. I wanted useful answers to become editable Notes and Documents. I wanted memory that I could search, review, correct, and delete. And I wanted to decide what context the assistant was allowed to use.
None of these things sounds revolutionary, and that is the point. AI is already useful. I was just tired of spending so much time operating it.
Lutin is still a work in progress, but I use it for the same reasons I started building it: less waiting, less switching, less searching, and much less copy and paste.
If you have the same annoyances, you can try Lutin for macOS.