Choosing AI tools

You're standing in line for coffee and the names keep coming. ChatGPT. OpenAI. Claude. Anthropic. Gemini. Google. Grok. xAI. Someone says one is the best now, someone else says that was last month, and nobody tells you how any of it fits together. It all sounds like the same thing in eight different hats.

It's not as confusing as it looks. There are only three layers, and once you see them you can stop pretending you missed a memo.

Companies

At the top are the labs. The firms racing each other to build the smartest engine on the planet, spending billions doing it. OpenAI. Google. Anthropic. xAI. They're the manufacturers. You will almost never log into one directly, the same way you don't drive to the factory to buy a car. They build the engine. Somebody else builds the thing you actually use.

Platforms

The platform is the app you open and log into. The thing with an icon on your phone and a password you forget. Each one is owned by a company. ChatGPT is OpenAI's. Gemini is Google's. Claude is Anthropic's. Grok is xAI's. This is the layer most people mean when they say "I tried AI." They opened one of these tabs and typed something into the box.

Models

Under the hood of every platform is a model. That's the engine. The model does the thinking, the platform is the car built around it, and a model always carries a version number.

Here's the part worth burning in. A newer model ships every few months. The labs leapfrog each other constantly, so whichever one is best today is not the one that's best by fall. By the time you've memorized this month's ranking, it's already changed. Chasing it is a full-time job, and it isn't yours.

What actually differs for an agent

The platforms do have personalities. Claude is known for writing in a notably human voice. ChatGPT is strong at reading long documents, a 30-page inspection report without losing the thread halfway down. Gemini lives right inside your Google account, next to your mail and your calendar.

That's the honest version. But for the four jobs you care about, drafting, reading, remembering, and looking things up, any current one is fine. The differences the experts argue about online live at edges an agent never touches. You're not training a model. You're trying to get a follow-up written before the option period closes.

So how do you choose

You don't choose a model. You choose a tool, and you let it worry about the model.

• Does it do a real job you have. Not a demo. A follow-up, a set of contract dates, a CMA you can hand a client.
• Does it bring your context to every task. A chatbot in another tab knows nothing about your deals, so you feed it everything by hand every time, which is just a typing exercise. The version that saves you time already has your inbox, calendar, transactions, and contacts, so it walks in knowing the deal.
• Does it live where you already work. Not a fifth tab you forget to open.
• Does it keep you in the loop. It drafts, you read, you send. Nothing leaves your name without your yes.
• Does it run a strong engine and swap it for a better one when that ships. So you never have to re-choose.

The model is plumbing. Your deals are the point.

What that looks like on a real deal

A newer engine ships on a Tuesday. An announcement, a thread of people arguing about benchmarks, a few breathless posts. You don't read any of it. Your tool quietly swaps the engine under the hood, and the next morning the follow-up on the Henderson deal is drafted in your voice, option period closing Thursday already noted, waiting for you to read it and send. You didn't pick a model. You just kept working.