If you run a business of 40 people and tried to follow AI news last week, you'd be forgiven for feeling a little seasick.

On 17 September, Anthropic reported that its model, Claude, now "leads" around 26% of the company's own research and development work. At the start of the year that figure was effectively zero. Headlines followed quickly, some of them framing it as an AI building its own successor.

Five days later, MIT Technology Review published an op-ed by researchers Timnit Gebru and Emily M. Bender titled "Don't be fooled by this summer of AI hype". They argue that several of the season's biggest capability claims look very different once independent experts examine them.

So which is it? Is AI about to rewrite how every business works, or is most of it marketing?

For a small or mid-sized business, the honest answer is that neither headline should decide anything for you.

Read the claim, not the headline

The Anthropic figure is more specific than much of the coverage suggested. According to Bloomberg's reporting, "leads" means Claude completes most of a task end-to-end from a high-level prompt, with a human supervisor overseeing the result. Anthropic also says Claude collaborates with staff on about 90% of their work. That's a meaningful shift in how one AI lab gets its own work done. It isn't the same as "AI is replacing researchers", and it certainly isn't the same as "AI is ready to run your finance team".

The critique has limits too. Scepticism about vendor claims is healthy, and Gebru and Bender are right that big announcements deserve scrutiny. But an argument that some headline claims are overstated doesn't show that the everyday tools are useless. Plenty of SMEs are already saving real hours on drafting, summarising and first-pass analysis.

Both things can be true at once. The capability is moving quickly, and a lot of what gets said about it is inflated. Most business leaders don't need to settle that debate. They need a way of working that doesn't depend on who wins it.

Why this matters more than it seems

The noise has a cost. ManpowerGroup's 2026 research found that confidence in AI adoption fell 18% year-on-year even as usage rose 13%. People are using AI more and trusting their own judgement about it less. When every week brings one story saying you're behind and another saying it's all hype, that's the predictable result.

Low confidence changes behaviour. Owners freeze, waiting for the dust to settle. Or they overcorrect, buying a platform because a competitor mentioned one. Neither decision comes from the work itself.

The only test that matters

Go Wisely's approach is simple. Judge AI by what it does in your workflow, not by what it does in a press release.

In practice, that means choosing one real, repetitive task this week. It might be a client proposal, a monthly report summary, or the weekly round of chasing emails. Run it through a capable assistant such as Claude, or ChatGPT or Copilot if that's what your team already uses. Then ask three plain questions. Did it save time? Did the quality hold up? What would a person still need to check?

That small experiment will tell you more about AI's value to your business than a month of reading the news. It also builds exactly the kind of judgement the ManpowerGroup figures suggest is slipping: the ability to decide for yourself what AI is good for, rather than borrowing a view from the loudest headline.

This is also why the Value dimension of Go Wisely's AI Accelerator Diagnostic asks "how will you know if any of this is working?" before it asks which tools to buy. A business that measures AI against its own work is hard to rattle, whether the headlines say yes or no.

A calmer way to follow the news

None of this means ignoring AI news. It means reading it with one question in mind: what exactly is being claimed, and would it change anything about how my team works on Tuesday afternoon? Most weeks the honest answer will be "not yet". When the answer is "yes", you'll know, because you'll have a real baseline to compare against.

The businesses that do well with AI over the next few years won't be the ones that react fastest to every announcement. They'll be the ones that kept testing it on their own work while everyone else was arguing about the headlines.

Go wisely.