AI assistants are supposed to save us time.
And often they do. You ask a question that might otherwise take ten minutes of searching, and a few seconds later you have an answer.
But there is a small problem with the way many of us use AI today: we have to leave what we are doing to use it.
You are working in Photoshop and cannot remember how to achieve a particular effect. You are editing a document and want a paragraph rewritten. You are looking at an error message and want to know what it means. You are writing an email and need a quick translation.
The question itself might take five seconds to ask.
Yet the process often looks like this: switch away from your application, find the browser, open an AI tab, click into the conversation, type the question, read the answer, then find your way back to the application where the question originated.
None of these steps is difficult.
That is exactly why the problem is easy to overlook.
The real cost is losing your train of thought
There has been plenty of discussion about whether AI makes people more productive. I think a more interesting question is where AI sits in the workflow.
When I am concentrated on a task, I generally do not want to change environments just to ask something.
If I am editing an image, I want to stay mentally inside that image. If I am working on a document, I want to remain focused on the document. If I am coding, I want to keep thinking about the code in front of me.
Opening another application is not a huge interruption in isolation. But it creates a small break in attention.
Sometimes that break is enough to notice another browser tab. Or an email. Or a notification. Or something you meant to look up earlier.
Five minutes later, the original five-second question has become a detour.
This is not really an AI problem. It is a context-switching problem.
Most AI assistants are still destinations
The dominant interface for conversational AI still treats the assistant as somewhere you go.
You open ChatGPT, Claude, Gemini or another service and enter its environment. That model made perfect sense when AI chat was something novel that we deliberately sat down to use.
But AI is increasingly becoming a utility.
I do not necessarily want to have an extended conversation every time I use it. Sometimes I just want to know:
- How do I do this in Photoshop?
- What does this error mean?
- Can you make this paragraph clearer?
- Translate this sentence.
- Summarize what I have selected.
- What PowerShell command do I need here?
These are tiny interactions.
The ideal experience is almost closer to using a calculator than opening another workspace. Ask the question, get the answer, continue working.
AI is more useful when it stays beside the work
I have increasingly found that the best place for an AI assistant is not necessarily full-screen.
It is beside whatever I am already doing.
A small assistant at the edge of the desktop changes the interaction surprisingly quickly. Instead of thinking, “I need to go ask AI about this,” it becomes closer to asking someone sitting next to you.
You can remain in the application you are using, ask the question, glance at the answer and carry on.
That difference sounds minor, but for short questions it changes the entire rhythm.
Consider the Photoshop example.
You already know how Photoshop works. You know roughly where everything is. You simply cannot remember the exact steps for one particular operation.
A traditional web search may lead to an article, a forum post or a ten-minute video. A general AI assistant can answer immediately, but you still have to leave Photoshop to have the conversation.
What I really want in that situation is simpler:
Stay in Photoshop. Ask. Read. Continue.
No new workspace. No hunting through tabs. No unnecessary interruption.
Not every AI question needs your most powerful model
There is another benefit to thinking about AI as a utility rather than a destination: you start thinking differently about the model itself.
For many everyday questions, I do not need the most expensive reasoning model available.
If I am asking how to crop something in an application, rewrite two sentences or explain a straightforward error message, a smaller and faster model may actually provide the better experience.
The response arrives quickly, costs very little, and I can continue with the task.
For more difficult work, I can switch to a more capable model.
This is one reason I like the idea of bring your own key, or BYOK, AI software.
BYOK gives the user back some control
A lot of software is moving toward the same commercial model: create an account, subscribe every month and use whichever AI models the product has decided to provide.
There is nothing inherently wrong with that approach. Bundled subscriptions are convenient for many people.
But they should not be the only way to use AI-powered software.
With a BYOK application, you provide your own API credentials and choose the provider or model you want to use. The software provides the interface and workflow, while the AI service remains your choice.
I find that separation appealing.
If a fast inexpensive model is sufficient for routine questions, use it. If you need a stronger model for a difficult task, switch. If you prefer another compatible provider, you are not necessarily locked into the application’s own AI subscription.
You are paying the AI provider for the AI you actually consume rather than automatically buying another recurring AI bundle simply because you wanted a useful desktop interface.
For people who already use APIs, it also avoids an increasingly common situation where you end up paying several different subscriptions for products that ultimately give you access to similar underlying models.
Local-first software also makes more sense for this kind of tool
If an AI assistant is going to sit beside my everyday work, I also prefer it to behave like a desktop application rather than another cloud account.
There is no reason every interface needs its own copy of my conversations, documents and settings stored on somebody else’s servers.
A local-first approach can keep application history and configuration on the PC while sending a prompt only to the AI provider the user has chosen when they actually ask something.
It is a refreshingly simple model:
- the desktop application handles the workflow;
- your computer holds the local application data;
- your chosen AI provider handles the AI request.
There are fewer layers between the user and the model.
And importantly, the software does not need to become another AI platform in order to be useful.
This eventually became a small Windows project
I started thinking about all of this because I kept encountering the same irritation in my own work.
I use AI constantly, but I did not always want to go to AI.
I wanted it nearby.
That eventually led me to build a Windows application around the idea.
The goal was deliberately simple: keep an AI assistant at the side of the desktop, let it stay above the application I am working in when useful, and let it retract to the edge when I do not need it.
Over time I added the things I personally wanted around that basic workflow: selected-text actions, screenshots, document input, local folder context, configurable AI providers and local conversation history.
But the feature I keep coming back to is still the simplest one.
I can ask a small question without leaving my work.
That has turned out to matter more to me than I expected.
Maybe AI needs to become less visible
There is a tendency in technology to make important products increasingly large and elaborate.
AI may move in the opposite direction.
As the models improve, perhaps the interface itself should disappear further into the background.
We may not need to spend more time inside dedicated AI applications. We may instead use AI in smaller moments throughout the applications where the actual work happens.
A question here. A rewrite there. An explanation when we get stuck. A quick second opinion without abandoning the thing in front of us.
The best AI interaction might eventually be the one you barely notice.
You ask.
You get the answer.
And you keep working.
If this way of working sounds useful, the Windows app I built around the idea is called SideNote Pro. You can learn more about using an AI assistant beside your work at sidenotepro.com.
