There’s a particular reaction around AI that I genuinely don’t understand.

Not criticism of AI. There are plenty of legitimate things to criticize.

I mean the reaction people have when they discover **which ** LLM someone else uses.

“Oh, you use ChatGPT? Gross.”

“Gemini? Cringe.”

“You still use Claude?”

Or the inverse version where somebody treats their preferred model like they just selected the correct faction in an RPG and everyone else has unfortunately chosen the idiot class.

I don’t get it lol.

Not because I think all models are equal. They obviously aren't.

I have preferences. Strong ones, actually. I use different models and products for different kinds of work because they behave differently, have different strengths, exist inside different ecosystems, and sometimes one just fits a particular workflow better.

But that’s precisely why the tribalism makes so little sense to me.

**A model is a tool. It is not an identity.**

And more importantly:

**Model ≠ system.**

That distinction gets lost constantly.

Someone sees a person using GPT and assumes they know the quality of that person's AI workflow.

They don't.

Someone says Claude is “better” than Gemini as though “better” exists independently of the job being performed.

It usually doesn't.

Better at what?

With what context?

Inside what harness?

Connected to what tools?

Using what instructions?

For a one-shot question or a persistent workflow?

Research? Coding? Writing? Analysis? Multimodal work? Google ecosystem integration? Agentic execution?

Those aren't footnotes. **Those are the system.**

You can put an incredible model inside a terrible workflow and get mediocre results.

You can also take a supposedly less impressive model, give it excellent context, strong tooling, clear boundaries, good retrieval, sensible verification, and a well-designed environment—and it can absolutely cook.

I think this is partly why the “ew, you use ChatGPT” thing feels so strange to me.

It sounds like making fun of somebody for using Firefox instead of Chrome without asking what they're actually doing on the computer.

Okay?

Maybe their setup works.

Maybe yours works.

Maybe both of you have entirely different requirements.

The useful conversation starts when we get past the brand name.

Tell me **why** you prefer Claude.

Tell me where Gemini consistently beats GPT for you.

Show me a coding workflow where one model produces noticeably better results.

Show me where another one falls apart.

Tell me about latency, tool use, context handling, reasoning behavior, hallucinations, cost, API ergonomics, ecosystem integration, personality, whatever.

Now we're talking about something real.

But “cringe”?

“Gross”?

What am I supposed to do with that information lmao.

It gets even stranger because frontier models leapfrog each other constantly.

Today's obvious winner becomes next quarter's “yeah, it's still pretty good.”

A model gets updated.

A provider changes a product.

A new harness exposes capabilities differently.

Someone discovers a prompting or context strategy that changes the comparison entirely.

The ground moves.

Building your identity around the leaderboard therefore seems like an exhausting way to use technology.

My approach is much simpler:

**Capability earns authority.**

If a tool performs well, I use it.

If another performs better for a particular job, I use that instead.

I don't need the first tool to become morally bad for the second tool to be useful.

ChatGPT can be excellent.

Gemini can be excellent.

Claude can be excellent.

And each can also annoy the absolute shit out of you in a completely unique and artisanal way.

That's software.

The more interesting question isn't:

**“Which model is best?”**

It's:

**“Which model is best here?”**

And eventually even that question expands into:

**“What combination of model, context, tools, workflow, and human judgment produces the best result?”**

That's the layer I care about.

Because once you're doing serious work with these systems, model fandom starts mattering a lot less than environment design.

The model supplies capability.

The surrounding system supplies context, memory, tools, constraints, verification, and execution.

The human supplies judgment.

Those are different jobs.

Collapsing all of them into “Claude good, GPT cringe” isn't technological discernment.

It's brand preference wearing a lab coat.

Use what works.

Change when it stops working.

Keep your standards higher than your loyalty.

The models certainly aren't loyal to you.