The race to crown one AI the winner may be missing the point. The operator is part of the system too.

People ask me some version of the same question all the time: which AI should I use? Usually they mean ChatGPT or Claude. Gemini comes up, and Grok, and Copilot if they work somewhere that runs on Microsoft. Lately Manus has joined the list. The question tends to arrive with a little fatigue attached, as though the person suspects a correct answer exists and they’ve fallen behind by not knowing it yet.

I understand why it feels that way. Every few weeks something gets announced as a breakthrough, a leaderboard reshuffles, and someone online declares that whatever you were using last month is finished. In that much noise, it’s reasonable to want someone you trust to name the winner so you can stop thinking about it.

I’m not going to name one. The longer I’ve lived with the question, the more I’ve come to think it rests on an assumption worth examining: that there is a single best system, and that a sensible person’s job is to find it and commit.

As far as I can tell, there is no one model to rule them all. At least not now.

We already live with more than one

Something occurred to me a while back while watching how entrepreneurs and small-business owners handle social media. Almost none of them believe they have to pick one platform and stay inside it forever.

A consultant might post regularly on LinkedIn, keep an Instagram account going, run a Facebook group for a local community, and check X when something is unfolding in real time. One of those usually gets most of their attention. That’s their home base. Another might sit mostly idle for months until some particular need makes it useful for a week.

Nobody finds this strange, and nobody accuses the consultant of disloyalty to LinkedIn. We understand, without ever having to say it out loud, that different digital environments can coexist, feel different, and earn their place in different combinations depending on who’s using them and why.

So I’m curious why AI seems to trigger the instinct to crown a single permanent winner.

Plenty of people get everything they need from one platform, social or otherwise, and that’s a perfectly good outcome. Having a home base just doesn’t obligate anyone to treat it as the only base.

The analogy breaks in one important way, and the break turns out to be useful. A social platform becomes valuable largely because of who is there. An AI environment can become valuable because of what you’ve built there: your history, your projects, your context, and your own learned sense of how to work inside it.

That difference explains why moving between AI systems can carry a kind of friction that opening another social app never does, and it’s a thread worth following.

What my own experience can and can’t tell you

I’ve been working actively with these systems since 2023, long enough to have watched several versions of the future arrive and quietly leave.

Most of that work has been on what I think of as the conversational side of the technology: prompting, reasoning through problems, writing, developing projects, learning how different models behave, getting output I can actually rely on, working within their limitations, and increasingly thinking about governance.

I’ve done some coding-related work with Codex and with Claude, and I’ve run into a good number of the newer agent tools along the way. I wouldn’t claim anything close to mastery of every agent, developer environment, API or skill system now on offer.

Over that time I’ve experimented across most of the major systems, sometimes extensively and sometimes only enough to understand what changes when I work somewhere else. I spend far more time in some than in others, ChatGPT most of all.

I’ve also developed ways of working across more than one system when the work calls for it.

What strikes me most, after several years of fairly intensive use, is that the landscape still expands faster than I can explore it. There are capabilities available right now that I haven’t seriously touched.

If that’s my situation, I’m skeptical that anyone can hand another person a permanent ranking and have it mean much.

My habits suit the kind of work I do and the way my mind happens to operate, and they aren’t meant as a template. Someone else may get everything they need from a single system and never think about it again, and I’d have no argument with them.

How a tool becomes useful

Very few people choose their AI by sitting down and evaluating every option. They start somewhere.

For some it was a friend demonstrating ChatGPT over dinner, or an employer that pays for Copilot so it’s simply there when the laptop opens. A company that runs on Google Workspace puts Gemini one click away. Others tried Claude because someone whose judgment they respect mentioned it, or drifted into Grok because they already spend their time on X, or used whatever was free and got a good answer early enough that they never had a reason to look elsewhere.

Those are the same reasons we end up with most of the tools in our lives, and I don’t think any of them are bad reasons.

What I find more interesting is what happens afterward.

You learn the environment. You figure out how to ask for things, where it tends to go wrong, how much context it needs, and which phrasing gets you what you want.

There’s a cynical reading of that process, which says people simply confuse the familiar with the superior. That probably happens sometimes.

The more useful observation runs the other way: familiarity changes how useful a tool actually is.

Hundreds of hours spent learning to work well inside one system build a real skill, and that skill belongs to the operator as much as to the software.

Familiarity is only the first layer. Over time a working environment fills up with projects, history and context that would have to be rebuilt somewhere else. Usage and context limits determine how much work you can get done before hitting a wall. Price changes the calculation too, especially when an employer is already paying for one system, has connected it to company tools and has approved it for company work.

For many people, that makes the company-provided option the genuinely right answer, not a fallback.

Then there are integrations, security requirements, and something as ordinary as whether you enjoy using the interface. A tool you find unpleasant is a tool you use less.

Put all of that together and a model that’s stronger on paper can still be the less useful choice.

Another system might outscore yours on some test and serve you worse on an ordinary workday, because moving the work means rebuilding context, adapting to a different environment, or paying for a tier you don’t need.

Nobody uses a model in the abstract.

What you experience as “an AI” arrives as a product: an interface, a price, a set of tools and limits, and more and more often an entire ecosystem with the model somewhere inside it.

That’s why I’ve become a little careful when someone tells me one AI is far better than another. I believe them, in the sense that I believe it was their experience. But I want to know what they were actually comparing.

It might have been a paid tier against a free one, a deep reasoning mode against a quick one, or a system holding months of their history against a blank slate. Two people can describe the “same” AI completely differently and both be reporting accurately.

None of this makes the models interchangeable.

Benchmarks exist for good reasons, and sometimes one system really is better at a particular kind of task. If your work leans heavily on that capability, it’s worth knowing.

A benchmark can tell me how one model performed against another on a particular test. It can’t tell me which environment I’ll find most useful for my work, under my constraints, at the price I’m willing to pay, given what I already know how to do.

That question involves me, and I usually only answer it by trying things.

Nothing here is holding still

There’s another assumption buried in “which is best”: that we’re choosing among stable products.

We aren’t.

The ChatGPT someone evaluated a year ago isn’t quite the ChatGPT they’re using now, and the same goes for every other name on the list. Capabilities, limits and prices all move, new tools appear between one month and the next, and sometimes whole categories of use seem to appear with them.

Some of these environments still feel mainly conversational. Some are woven into the office software people use all day. Some have become serious workspaces for writing, research, analysis or code. And some, Manus being the one that recently entered my own thinking, behave less like something you talk to and more like something you hand work to and check back on later.

The boundaries between those keep shifting, often within the same product.

The companies behind these systems don’t sort neatly either. Several of the largest are investors, infrastructure providers or partners to one another while also competing directly.

I’m not going to predict how that resolves, and I don’t know.

I just see no reason to assume that today’s lineup of products, or even of companies, will look the same two years from now. We are making permanent-sounding judgments about things that are nowhere near permanent.

Seen that way, maybe you aren’t really choosing an AI at all.

You may simply be choosing where to work, for now.

The operator

“Which machine wins?” is a question about the machine.

What I care about is what a particular person can accomplish with it, and that’s where everything above comes together.

Familiarity, budget, whatever access an employer provides, free versus paid, accumulated context, skill and above all purpose all shape the answer.

It’s also why two people can use exactly the same AI and walk away with wildly different opinions about how capable it is. Both of them were measuring a system with two parts, and only one of those parts was the model.

The other was them.

So the question I’d leave you with is smaller and harder than the one you came in with:

Which intelligence is most useful to you, for what you’re actually trying to accomplish, right now?

I don’t know the answer for you. It may be the one you’re already using, or something you’ve never tried, or more than one.

That’s yours to work out.

And choosing where to work is only the first question. Once the work starts coming back, a harder one follows:

How do you decide what to trust?

That one can wait.

Before you go looking for a winner, though, it may be worth asking why you’re using the one you’re using now.

Frequently Asked Questions

Which AI is best?

There probably isn’t one answer that applies to everyone. ChatGPT, Claude, Gemini, Grok, Copilot, Manus and other systems have different capabilities, tools, limits, pricing and ecosystems. The more useful question is which system works best for what you are trying to accomplish, under the circumstances in which you actually work.

Do I need to use more than one AI?

No. Some people may get everything they need from a single platform. Others may find value in using more than one. Having a primary AI or “home base” does not necessarily mean you have to use it exclusively.

Is ChatGPT better than Claude, Gemini or Grok?

That depends on what you are comparing and what you are trying to accomplish. Different models may perform better on particular tasks, but the model itself is only part of the experience. Subscription level, available tools, accumulated context, usage limits, integrations, familiarity and cost can all affect which system is more useful to a particular person.

Does paying for an AI make a difference?

It can. Free and paid versions may provide different models, tools, usage limits, reasoning capabilities and other features. That means two people who say they tested the same AI may actually have experienced meaningfully different versions of the product.

Should I switch AI platforms if another model performs better on benchmarks?

Not necessarily. Benchmarks can provide useful information about model performance, but they don’t account for everything that affects real-world usefulness. If you have developed considerable skill, context and workflow inside one environment, those things have value too. A benchmark can inform the decision without making the decision for you.

Can I use more than one AI at the same time?

There is nothing that requires you to treat AI platforms as mutually exclusive. Some people choose one environment and stay there. Others use different systems depending on their work, available tools, constraints or preferences. How many systems make sense depends on what you are actually trying to accomplish.

Why does familiarity with an AI matter?

Because the operator learns too. Over time you become better at framing requests, recognizing weak answers, supplying the right context and understanding how a particular system behaves. That experience can make a tool more useful to you even when another model performs better on a particular benchmark.

What does “Super Intelligence” mean in this article?

“Super Intelligence” has a more specific meaning in technical discussions than the AI systems most people currently use. I use the phrase here deliberately because it has entered the broader public conversation around increasingly capable AI systems. This article is about how we choose and work with those systems, not a claim that today’s consumer AI meets the technical definition of superintelligence.