Hello and welcome to Eye on AI. In this edition…OpenAI wins a $200 million Pentagon contract…Salesforce finds AI models can’t use CRM software very well…and a new study shows how AI scrapers are overwhelming cultural institutions.
So what’s AI got to do with this? Well, everyone is hoping that AI will save us from this perfect storm of impossible expectations meeting human limitations. But the technology itself won’t do this. In fact, a lot of the ways companies are deploying AI and people are using the technology could make things worse.
“AI is delivering real productivity gains, but it’s not enough,” Teevan tells me. “The speed of business is still outpacing the way we work today.”
She says that crafting prompts for AI to perform tasks for us, such as conducting research or generating a business presentation, “actually increases our metacognitive burden.” In other words, to write a good prompt, a person has to think clearly about the steps they want the AI to perform, and provide a list of dos and don’ts. This thinking process necessitates concentration, and it also requires someone to transform things they know tacitly into explicit instructions. Having to do this, “can feel overwhelming,” Teevan says.
But there are better ways to work with AI that can alleviate this burden—or at least share it. AI itself can be used to help craft prompts, for instance, Teevan says. Cambon says that too many people are viewing AI as just another software tool. It’s better, she says, to think about it like a digital colleague—something to which you can assign entire tasks or processes.
More importantly, to get the most out of AI, companies need to change their organizational structures, the way their employees work, and also how they measure value. Microsoft has identified companies they call “Frontier Firms” that are doing this. At these organizations, 71% of workers say their company is thriving, compared to just 37% globally.
Now, it should be said, there are not too many of these Frontier Firms out there. Out of 31,000 companies Microsoft looked at, only 840 met the criteria. Most of these companies were in tech—many of them so-called “AI native” startups that have the benefit of being able to design their processes around AI from the start. “They don’t have to unlearn a whole load of stuff,” Cambon says. But interestingly, she says that some of the Frontier Firms were in professional services, like consulting, accounting, and law, which is an area where AI is rapidly disrupting traditional work processes and even challenging entire business models.
For non-AI native companies, getting the full benefits of AI means changing organizational management and structures. “It is about how do you externalize knowledge and make things available for AI to learn from,” Teevan says. “It is about creating feedback loops and being very intentional about the content we create for our teams.”
Microsoft’s research suggests there are some key changes that differentiate the Frontier Firms from the rest. They prioritize impact over activity, focusing on the 20% of tasks that create 80% of a business’s value. They redesign workflows instead of just trying to automate them. (Rather than have AI write status reports, for instance, ask whether you need status reports in the first place.) And they increasingly use AI as agents that can handle entire workflows, not just individual tasks. In this world, employees become “agent bosses,” Microsoft says.
Cambon says that the Frontier Firms also tend to have much flatter organizational structures, where teams are organized around completing a specific project, not around areas of expertise. Does Microsoft have an interest in selling this narrative in order to convince companies to buy its AI software and cloud services? Sure it does. But that doesn’t mean it’s wrong. It is clear that the companies that get this right will have a big advantage. And the ones that don’t? They’ll just have increasingly efficient chaos and burnt out employees.
With that, here’s the rest of today’s AI news.
