Why We Don’t Have a Favorite AI Tool

One of the questions I hear most often from business leaders is some version of “Which AI is best?” Sometimes it comes out as “Should we be using Copilot, Claude or ChatGPT?” and sometimes it is simply “What’s your favorite?” Either way, the person asking is usually hoping for a single name so they can stop worrying about the decision.

I understand the impulse. New AI announcements seem to arrive every week, and no one running a business has time to evaluate all of them. But my honest answer is that we don’t have a favorite AI tool. After a great deal of real work with several of them, I’ve come to believe that “Which AI is best?” may be the wrong question to start with.

How the Work Actually Gets Done

I have spent an enormous amount of time working with Microsoft Copilot. I’ve used it to think through County Quest’s business strategy and consulting philosophy, to develop courses, to work through client situations, to shape our podcast, and to brainstorm ideas until they finally make sense. A considerable amount of context has built up in those conversations, and that context has real value to me.

More recently, I’ve also been taking certain kinds of work to Claude. Part of the reason is that I like having a second angle applied to my ideas. Part of it is practical: much of my finished work ends up as Word documents, and Claude does a very nice job of producing finished documents. That doesn’t make Claude the best tool for documents, and it won’t necessarily stay that way. It simply works for me right now.

It would be easy to boil this down to “Copilot does the thinking and Claude does the writing,” but that would be the wrong lesson. What actually happened is that I gradually learned how I like to work with each of them, much the way you learn how you like to work with different people.

This article is an example. The idea, the argument and the structure were developed in conversations with Copilot, which then helped me put together a careful handoff package so Claude could take the work into its next stage. No AI had to win. The work simply moved to the environment that made sense for the next assignment.

Think of a Good Team, Not a Tournament

Picture a business with several capable, conscientious employees. A good manager doesn’t hand every assignment to the same person just because that person is dependable, and doesn’t spend time trying to crown the single best employee either.

Instead, the manager gets to know the team. One person thrives on untangling an ambiguous problem and figuring out what is really going on. Another is excellent at turning a pile of information into something polished. Someone else lights up when there is research to do, numbers to work through or detailed technical work. Over time, assignments find their way to the people who suit them. Nobody needs to hold tryouts every time a new task comes along.

AI tools can be approached in much the same way.

Trying a New Tool Doesn’t Require a Project

My own move to Claude was far less dramatic than it might sound. I had been hearing good things about it, so I tried a subscription. I didn’t launch a formal evaluation or spend weeks studying it. I gave it a few low-value tasks that could be done quickly and compared the results with what I was used to after about two and a half years of working deeply with Copilot. Those small, low-risk assignments told me more than reading reviews ever would have, and over time I figured out where it fit in my work.

This is not a recommendation that anyone become an expert on Copilot, Claude, ChatGPT, Grok and half a dozen others so they know which one is best for every possible task. That would defeat the purpose. Many of the leading general-purpose AI tools do a perfectly acceptable job across a remarkably broad range of ordinary business work. Sometimes there is a meaningful difference between them, sometimes there isn’t, and sometimes the AI you already have is more than good enough. Spending three hours deciding which AI should handle a thirty-minute job is rarely a good trade.

What we recommend is refreshingly simple: start using it, experiment and pay attention to what works. Give AI real work, the kind that is actually sitting on someone’s desk this week, and let your people experiment under sensible rules about security and how company information is handled.

Over time, they will form their own observations. One person will find a particular tool better for thinking through something complicated. Another will notice that a different tool handles long documents especially well, or that Copilot is handy because it works with information already in Microsoft 365. Sometimes the honest observation will be that, for a particular job, either one works just fine.

There is some relief in this. Letting one or two curious employees try another paid AI product for a couple of months is an inexpensive experiment compared with most technology projects. It doesn’t necessarily require a committee, a six-month evaluation or a giant AI strategy before someone finds out whether a tool is useful.

Inexpensive Doesn’t Mean Careless

There is another side to this. Company information deserves the same respect it always has. Any experimentation should happen within your security, privacy and confidentiality policies, and if those policies don’t exist yet, that conversation should come first. A small experiment with real work and responsible boundaries is a very different thing from buying every AI product on the market and seeing what happens.

I would also avoid handing out permanent job titles, with one tool as the writer, another as the thinker and a third as the researcher. We have our own preferences, but they aren’t laws of nature, and these products change so often that what is true this fall may be out of date by spring. Different people also work differently with the same AI. One employee may build a terrific working relationship with a tool that a colleague down the hall can’t stand. The useful question is less about what the internet says a tool is good at and more about what you’ve learned it does well when it is working with you.

Four Bigger Lessons

Your context has value. When you work in an AI environment for a long time, it accumulates your terminology, preferences and philosophies, along with decisions you’ve already made and ideas you’ve already rejected. Walking away from one of those environments completely would carry a real switching cost, because so much of your thinking already lives there.

Your information needs a home. This is where the first lesson can lead you astray. The answer is not to pick one AI chat and turn it into your company’s permanent memory. A final procedure, a client deliverable, a policy or an important decision should each have an intentional, authoritative home, and AI should help people work with that information without being the only place it lives. The AI doesn’t necessarily need a permanent home. Your information does.

Organization matters more, not less. The future probably isn’t one magical AI that does everything. More likely, you will have several AI systems, a few agents and AI built into the tools you already use. That is wonderful, but powerful AI pointed at disorganized information and fuzzy responsibilities can simply help you build a more modern version of the same chaos.

Friction comes before features. The question shouldn’t be “We bought Copilot, so what can we make people do with it?” or “Everyone is talking about Claude, so what should we use it for?” Start with what you are trying to accomplish and what is getting in the way. Then decide whether AI can help and which of your tools is a reasonable fit. Quite often, the one you already have will do just fine.

The Better Question

The most important question may not be whether your company should use Copilot, Claude, ChatGPT or whatever comes along next. It may be whether your business is organized well enough that whatever AI comes next can actually help you.

A practical place to start: this week, take one real task from your list, give it to the AI you already have and pay attention to how it goes. If someone on your team is curious about another tool, let them try it on real work within your rules, and ask them in a month what they noticed.

Keep your important information where it belongs, start with the friction, and let the tools earn their assignments over time.

If you would like help seeing where the friction is in your own business, the free Friction Finder survey at buildacalmerbusiness.com is a good first step.