I split Claude, Gemini, and ChatGPT into work, home, and casual, and it's a mixed bag

The explosion of LLMs, while intended to simplify our lives, has actually created a new flavor of chaos. Dozens of AI browser tabs and app windows crawl over my screen, whether I’m using a computer, a tablet, or a phone. To make matters worse, I admit that I can’t live without them now. 

I’ve long tried to leverage the collective versatility of AI models by continuously hopping between them. This approach got my work done, but at the cost of messy conversation boxes and occasional miscommunication. I didn’t blame any of them for such inconveniences — I contaminated their context windows myself with all kinds of questions, from work queries to sudden breezes of thought.

It’s time to bring it under control.

I spent some days with each of the three models assigned for fixed use cases — Claude for work, Google Gemini for home, and ChatGPT for casual quick prompts. Did that end the chaos or create a new source of trouble?

The honeymoon phase

A quick word on my personal context and daily demands to see if your workflow looks anything like mine. I engage in quite diverse work: language translation, writing, image processing, long-document reading, and light coding — all of which are allocated to Claude. I also operate a DIY smart home with a voice-controlled interface on Gemini, while ChatGPT handles impromptu brainstorming.

This mindful separation seemed to be a cure-all for my problems during the first days. Each AI worked as a specialized assistant with distinct roles, eliminating my tendency to tap whichever was within reach. I knew exactly where to go in a specific setting and which tone to use in the prompts to get the corresponding answers.

My workflow just flowed. Most work sessions with Claude were uninterrupted, enabling me to make further progress on each task and finally finish end-to-end projects with results that exceeded my initial expectations. The ultimate payoff of the project is a custom web dashboard that automatically tracks my published articles and scores them for conciseness, tech-savviness, practicality, and more. The evaluation is plotted on a clean, straightforward editorial radar for future iteration. Credits go to the Claude-driven deep work — isn’t it a nice way to deal with my shortened attention span these days?

Moving on to home tasks with Gemini, I’ve been using hands-free home control with it for weeks now, and offloading work tasks from the AI eased the burden it’d shouldered for the better part of the year. Repetitive home-control gestures on such a familiar interface felt more like second nature than ever. It’s an always-on voice hub and better suited to the task than Claude’s text-heavy interface.

The deep integration of Gemini into my Google ecosystem, which includes a Google Pixel phone, Google Home, Calendar, Keep, and Gmail, turned entering prompts and immediately having my needs met into a streamlined, clutter-free process. This experience draws me to interact with my smart home setup more frequently.

Although ChatGPT appears to play a less essential role in this project, its absence would’ve caused noticeable disarray. It’s a Swiss Army knife for throwing ideas around on the fly without cluttering the other two’s long-term context windows. The sheer frequency of me opening ChatGPT revealed how fragmented my thinking was: I just couldn’t let these sparks of thought go away; they could be great sources for future projects. If you want to improve your focus, this setup is likely for you: resist the urge to open your scratchpad and instead double down on your primary work AI.

Just kidding — a digital detox isn’t why I delegated everyday queries to ChatGPT. I came to it with all kinds of random questions you can imagine: where to take my son on the weekend, how to enhance my look in a photo, or how often my cats should shower. Honestly, this is no less important than my work or home tasks; getting such everyday information is the way many of my friends leverage their AI. Bombarding it with a series of random questions gradually built its understanding of my lifestyle and even of me as a person outside work. Alongside my consistent double-checking and fact-checking, it’s an indispensable part of the setup.

Last but not least, I could track my token allocation across the three models to statistically assess whether my day was primarily driven by work productivity, deep learning, or personal connection. At first, I thought this tracking also enabled token optimization and saved me some money, but things didn’t turn out the way they should have. Buckle up before we jump into the downsides.

There’s more than one catch

A harsh truth soon surfaced: assigning one AI all tasks of the same category inadvertently lets others’ strengths go to waste. The literary richness in Gemini’s suggestions was absent throughout the experiment. As a writer, I really missed its expressive flair (without any offense to Claude’s nuanced and sophisticated writing).

More than that, when Gemini Spark is available for more users, it’ll certainly be a useful background automation engine in their workflow — drafting email responses and extracting Calendar entries at a bare minimum. Limiting its functionality to generating drafts rather than responding to an email right away sets a clear boundary, allowing me to proofread and remain the final decider of how my replies look. I can’t let such a nice feature slip by just because I’ve chosen Claude as the primary work AI. In short, this setup lacks longevity and may prove fragile in the face of any new features or updates from any of the three models.

Not to mention the token allocation dilemma. My immense workload quickly drained the Claude tokens, whereas the other two still had plenty to spare.

Of course, I could switch between Claude and Gemini for different tasks, but it would undermine the clear separation established from the start. My workflow may become fragmented again, but at that stage, I accepted that some fragmentation was even necessary to maximize the individual potential of these AI models.

Extreme separation deprives AI models of the full picture, which proves especially valuable in hybrid tasks — my home data often provides useful input for work, and vice versa. For instance, I view troubleshooting my smart home automation setup as a work task because it can be an excellent material for my tech articles. However, I have to transfer data and context from Gemini (my home hub) to Claude (my work AI) to do this, and it’s clearly another layer of friction. Operating directly within Gemini is way more straightforward.

While I use fewer words for this downside section, these caveats are by no means trivial. They are critical friction points in a superficially neat setup that should have worked seamlessly under my complete control.

The experiment didn’t go to waste

A complete separation is only possible when one’s personal and professional contexts are clear-cut. As a remote worker, I find that my setup has created a work-from-home paradox. I can’t stop controlling my household appliances through Gemini just because I’m in the middle of an article or a coding process with Claude. It’s great to have them in separate interfaces doing different things, but this means the number of tabs/apps open remains the same, and my attention still snaps — just less frequently.

Of course, I took things to extremes in this setup — using which AI model for what purposes should come down to individual preferences. A complete separation setup may not be for me, but it’s unfair to say the experiment does no good at all. It tidies up my workflow and clearly shows where it breaks down. From there, I’ve decided to switch from a complete division to a flexible hybrid approach — allocating workloads based on each AI’s strengths rather than partitioning them by life domains to leverage the token limits of all three. 

Now I lean on Gemini for home automation and writing assistance (angle testing and data summarization). Meanwhile, Claude is responsible for logical tasks and coding, and ChatGPT remains a readily available host for random thoughts. I can now hop between them without feeling guilty. Using only one is already complicated enough, as I need to keep each chat box focused on a single context for streamlined answers, but embracing some compromises and an easygoing view towards fragmentation makes things easier. It’s something that we do more often than not, for our own phones, operating systems, and DIY projects.