I just saw this article on The Verge:

Let me start by saying that I am not a fan of chatbots as the primary user interface for enterprise systems. Although LLM-backed tools are significantly better at detecting intent and making data more useful, as a survivor of the fever dream that was Chatops in the mid-2010s, I don’t see how they solve any of the fundamental issues we faced with the previous iteration. Problems with permissions, concurrency, versioning, reuse, verbosity, and error handling are all still present, and we now have to deal with fresh new challenges brought by AI systems, such as hallucinations.

Even if these problems were to be solved, I believe that we tend to overestimate our desire to talk to machines. Conversational interfaces can be great when you are exploring something unfamiliar, don’t have a device handy, or when you want to ask questions. However, they are inefficient and annoying when you are on a laptop or phone and know exactly what you want, from turning off the lights to creating a pivot table.

That’s why Outropy doesn’t have a chatbot. Instead of asking users to talk to our system, Outropy offers a user experience inspired by tools such as VSCode and Intellij. These integrated environment make coders highly productive by providing timely and relevant insight and allowing users to “double-click” for contextual and actionable information about anything they see on the screen.

To our dismay, Microsoft invested millions of dollars in marketing to convince the world that copilot is just a synonym for chatbot. That’s why we stopped referring to Outropy as the Copilot for Engineering Leaders and started calling it The VSCode for everything else.

But, back to Microsoft, it’s no secret that the adoption of Copilot for both Windows and Office has faced issues. I suspect that they are facing the same issue as many AI startups: a lot of users sign up to try the product upon launch, investors get excited about massive revenue projections, but users start to churn as soon as the first or second bill arrives. It’s easy to get IT managers who need to provide some kind of “AI strategy” to their bosses to sign up for a trial, but once they’ve played around with it for a little while it’s hard for them to justify what the extra $30/user/month buys them.

47e3653 (copilotcalypse)

Anyone paying close attention to the enterprise AI market could see this copilotcalypse coming, but I had hoped that it would be a forcing function pushing Microsoft and other deep-pocketed companies who are haphazardly plastering ✨ buttons all over their systems to invest in some UX research to move away from chatbots.

Instead, they are doubling down and adding new chatbots to help you extract value from their existing chatbots. Chatbots will keep being added until engagement metrics improve.

This is disappointing, but not surprising. I have worked in tech for long enough to know that inside a large organization such as Microsoft, Google, or Meta there are so many competing goals that the safest way to keep your head on your neck is always to keep doing what the higher-ups like, even if it’s clearly not working.

And this is the exact kind of opening that startups exploit to take on incumbents.