When Do I Prefer Using Different AI Tools?
There are so many AI tools in the market now, and every few weeks there seems to be another model, another coding agent, another research assistant, or another creative tool that promises to do everything better than the previous one.
Because of that, I have stopped thinking too much about which AI is the best and started thinking more about which AI makes sense for the problem I am trying to solve.
I don’t actually use one AI for everything.
Different tools have started to feel like different kinds of coworkers, where the tool I choose depends largely on whether I already know what I want, whether I need help figuring it out, whether I need factual information, or whether I simply want to explore an idea through conversation.
Here is how I personally think about it.
Claude → When I already have a proper plan and need to build it
When I already know what I want to build and the task involves a significant amount of code, I usually prefer Claude.
For example, if I already know that I want a Spring Boot backend with PostgreSQL, JWT authentication, an Angular frontend, a particular API structure, and a specific architecture, I don’t necessarily want the AI to spend most of its effort deciding what the project should be.
I already have the direction.
What I need is help turning that direction into a working implementation while keeping the architecture, code structure, and requirements consistent.
That is where Claude fits naturally into my workflow.
I can give it the architecture, requirements, constraints, existing code, and expected behaviour, and then work through the implementation with it.
So my mental model is:
I already have the plan, and Claude helps me execute the plan.
This doesn’t mean I think Claude is automatically better at every coding task, because the capabilities of these tools change constantly. It simply means that I personally find it particularly useful when I have already done the higher-level thinking and want an AI to help me move from that plan toward implementation.
Antigravity → When I know the goal but don’t have a strong plan yet
My workflow becomes different when I know what I want to achieve but don’t really know what the final implementation should look like.
Sometimes I might have a rough idea such as wanting to build an application, automate a particular workflow, or solve a technical problem, but I don’t yet know which architecture, files, tools, or sequence of actions will actually get me there.
In that situation, I don’t necessarily want to hand the AI a perfectly defined plan because I don’t have one myself.
I want to figure it out along the way.
That is where I prefer an agentic environment such as Antigravity.
Instead of approaching the interaction as:
“Here is exactly what I want you to code.”
I can approach it more like:
“Here is what I am trying to accomplish, so let’s figure out how to get there.”
That distinction makes a surprisingly big difference in how I use AI.
When I already know the solution, I want execution.
When I don’t know the solution yet, I want exploration.
So my personal mental model is:
Claude is closer to execution of a plan, while Antigravity is closer to exploration toward a solution.
The distinction is not permanent, though, because AI products are rapidly absorbing each other’s capabilities, and the boundaries between coding assistants, agents, research tools, and general-purpose assistants are becoming increasingly blurred.
ChatGPT → When I just want to talk through something
ChatGPT is probably the simplest category for me because there are many situations where I don’t need a specialised workflow at all.
Sometimes I have a question, an idea, a random problem, something I want to understand better, or even a thought that I haven’t completely formulated yet.
In those situations, I don’t necessarily want to open a coding environment or start a research workflow.
I just want to talk through it.
That is when I naturally reach for ChatGPT.
I can start with one question, continue asking follow-up questions, change the direction of the conversation, challenge an answer, and gradually figure out what I actually wanted to understand in the first place.
For me, that is one of the most useful characteristics of conversational AI because the interaction doesn’t always have to begin with a perfectly defined objective.
Sometimes thinking through the conversation is the objective itself.
Gemini + Veo → When I need a quick AI-generated reel
This one is much more straightforward for me because sometimes I have an idea for a visual or a short reel and I don’t want to spend a huge amount of time setting up an entire video-production workflow.
I want to take the idea, describe what I want, generate something, make whatever small edits are necessary, and move on.
That is where Gemini and Veo fit into my workflow.
For example, if I have an idea for a social media post and want a visual representation of it, generative video can reduce the distance between having the idea and actually seeing something on screen.
My workflow becomes something like:
Idea → prompt → video → edit → publish.
I am not necessarily trying to produce a cinematic masterpiece every time, because sometimes I simply need a visual that communicates an idea quickly enough to use in a reel.
In those situations, the speed from idea to usable visual matters more to me than having an extremely complicated production workflow.
Gemini / Perplexity → When I need factual or current information
This is where my behaviour changes quite a lot.
If I am asking something like what happened recently, what the current price of something is, what the latest development is, or what a company recently announced, I don’t want to rely purely on what an AI model remembers from its training.
I want information that I can actually check.
That is when I prefer tools that are strongly oriented around web search, current information, and sources.
Perplexity naturally fits this type of workflow because the product is heavily focused on search and presenting information alongside sources.
Gemini is also useful for this kind of task, particularly when the question naturally connects with Google’s broader search ecosystem.
For me, the distinction is therefore less about believing that one particular AI is always factually correct and more about choosing a workflow where verification is built into the process.
If I am asking for a creative idea, I don’t necessarily need the answer to come with ten sources.
Get Veda’s stories in your inbox
Join Medium for free to get updates from this writer.
If I am asking about a current event, a technical specification, a company announcement, or a factual claim that matters, I care much more about where the information came from and whether I can verify it.
So this is roughly how my AI workflow looks
What I am trying to doWhat I usually reach forI already have a clear coding plan and want to execute itClaudeI know the goal but don’t have a strong implementation planAntigravityI want to have a conversation, brainstorm, or think through somethingChatGPTI need to quickly generate an AI video or reelGemini + VeoI need current or factual information that I can verifyGemini / Perplexity
But there is something important about this table.
These categories are not fixed, because every major AI company is trying to expand into everything.
Claude can search.
ChatGPT can research.
Gemini can code.
Perplexity can write.
Coding agents can browse the web.
AI assistants can generate images and videos.
The boundaries that existed between these products a year ago are becoming much less obvious.
I don’t think the future is one AI for everything
The more I use different AI tools, the less I think the future will be about everyone choosing one AI and using it for every task.
I think it will become much more personal.
Someone might use Claude heavily for software development, ChatGPT for everyday conversations, Gemini for information and Google’s ecosystem, Perplexity for research, and another specialised tool for design or video.
Someone else might have a completely different combination.
The important thing is not necessarily knowing every AI product in the market.
It is understanding what kind of problem you are solving and what kind of AI workflow that problem requires.
Maybe that is the real AI skill
Knowing how to write a good prompt is useful, but I think another skill is becoming increasingly important: knowing when to use which AI tool.
If I have a well-defined architecture, I want an AI that can help me execute it.
If I have only a vague objective, I want an environment that can help me explore the solution.
If I simply have a question or an idea, I want a conversational assistant.
If I need a visual, I want a generative media tool.
And if I need factual or current information, I want a workflow where sources and verification are part of the process.
So I don’t necessarily ask myself:
“Which AI is the best?”
I ask:
“What am I actually trying to accomplish?”
Then I choose the tool.
And that small change in thinking has made using AI much more useful for me.
And sometimes the tool you need isn’t another AI
There is one more thing I have started thinking about when it comes to productivity.
Not every problem needs another tool, another app, another prompt, or another AI workflow.
Sometimes the best thing you can do is simply pause for a moment.
When you are constantly switching between tabs, messages, coding environments, meetings, notifications, research, and AI conversations, it can be surprisingly easy to keep working without ever actually taking a proper pause.
That is where breathing can become a simple tool to have in your everyday routine.
The interesting thing is that there isn’t necessarily one breathing technique that everyone will naturally prefer.
Some people like slower breathing.
Some prefer counting.
Some prefer a very short breathing exercise they can do between tasks.
Some prefer something they can use before focusing.
Others simply want a simple technique for creating a small pause during a busy day.
So instead of giving you one generic breathing exercise and saying “this is the one you should use,” I created something more practical.
Find Your Breathing Technique Guide
It is a simple guide designed to help you explore different breathing techniques and find one that fits the kind of pause you are looking for.
Inside, you can explore techniques based on situations such as:
- when you want to slow down for a few minutes
- when you want a simple reset between tasks
- when you want to create a focused pause before starting work
- when you want a short breathing practice that fits into a busy day
- when you want to experiment with different breathing patterns and see what feels comfortable for you
The idea is similar to what I described throughout this article.
Don’t start with the tool. Start with what you need.
Just like I don’t ask “Which AI is the best?” before figuring out what I am trying to accomplish, I don’t think you necessarily need to find one “perfect” breathing technique.
You can understand the options, try them, and find what fits your own routine.
If you want to explore different techniques and find a simple breathing practice for your everyday routine, I made this guide for you:
Find Your Breathing Technique Guide → Get it here
A small pause can sometimes be more useful than another productivity tool.
And after all the AI tools, prompts, agents, research, coding, and automation, sometimes the most useful thing you can do is simply take a breath.
