Place Your Bets

Picking an AI platform is starting to feel like a game of chance

“You’re right that Claude has gotten dumb.”

This was the text I got from my one of my colleagues yesterday. She was following up on a conversation we had earlier in the day where I was lamenting the battles I had with Claude Cowork over the weekend. Tasks that I thought were locked and loaded – that Claude could handle without me hand-holding – had suddenly become unreliable and it was producing things that were just plain wrong.

I use Claude Cowork to help me curate and draft my weekly newsletters. I feed it a laundry list of links to stories I’ve collected throughout the week and ask it to help me organize everything. The process had become seamless in recent weeks, but last weekend was different. Claude replaced actual URLs I had given it with made-up links, causing rampant 404 errors. It straight-up missed several articles I had shared. And it kept delivering the drafted text in HTML instead of the usual Google Docs.

I had flashbacks to last year when I had similar battles with ChatGPT – and which ultimately led to me switching most of my work to Claude. Was Claude now becoming as unpredictable and infuriating as MattGPT? Sure looked like it.

“It’s literally changing by the minute”

Business owners face a real challenge right now. They know they need to “adopt AI” but aren’t sure exactly what that means in the context of technologies that have built their businesses on shipping new models every week.

I heard this explicitly on a call with an agency owner recently. We were talking about her company’s AI adoption and she shared that they were “in the middle of transitioning everything to Claude," but also, "I have a little question mark in the back of my head because some things are feeling easier in ChatGPT, but it's literally changing by the minute."

She’s also getting barraged with upgrade requests from the other tools in her tech stack – everyone coming with hands out to pay for the AI upgrades they’re implementing, even if many of them are AI vaporware.

So how do you build a business on a foundation that’s constantly shifting underneath you? I have colleagues and clients who have platformed on ChatGPT, but what happens when OpenAI releases its new “superapp” that will supposedly kill the chat interface that we’ve all become accustomed to? Agency owners I know have built complex workflows in Claude, but what do you do when it becomes unstable like it did for me last weekend? And was that instability caused by the imminent release of Fable, their new “Mythos-class” model.

Half on Red, Half on Black?

So is the answer – like many business owners are currently doing – to simply hedge your bets and invest in multiple models? Perhaps, but the costs are significant. We recently did an analysis for a client looking to give every one of their 200 employees access to both ChatGPT and Claude. The price tag? $240,000…per year. And that didn’t include the cost of extra compute, which recently cost one company $500 million in Claude token costs in a single month.

And the operational costs are enormous. How do you build repeatable processes, training and governance when you’re using two (or more) different tools – especially when the companies behind them are positioning themselves for trillion-dollar IPOs so need all the revenue they can get?

All of this see-sawing is also leading to employee burnout, with a recent Harvard Business Review study finding that 1 in 7 employees is showing signs of “brain fry” from juggling AI tools at work.

In short, hedging isn’t a workable AI strategy.

The Model isn’t the Answer

So what’s a business owner to do? Here's my honest take: don't build your workflows on top of raw model access.

Pick a primary model for now — the one that fits your team's work best today — but build your real competitive advantage one layer above that. Invest in the workflows, the knowledge systems, the governance, and the human judgment that sit on top of whatever model you're using. Those things travel. They'll survive the next model update, the next pricing change, the next time OpenAI rewrites its interface.

The model is infrastructure. Infrastructure changes. What you build on it is yours.

The ground beneath you won’t stop shifting anytime soon. Get used to moving your feet and finding balance again and again because it's not about the model. It’s about building something that doesn't have to start over every time the ground shifts.