Artificial Intelligence – Agent Debate

Great in some places, not in others

• Microsoft is making the case to turn the PC into an agent-driven user experience, but the reality remains that agents are going to be fantastic in some settings and very average in others.
• Microsoft will be holding its Ignite conference next week, where its key proposition will be Windows’ evolution into an agentic OS.
• This statement by Pavan Davuluri (see here) was met with derision, with the main theme of replies stating that an agent running Windows is something that they do not want.
• As a Windows user, I generally agree with this sentiment, as I have a Qualcomm-powered AI PC with CoPilot installed and generally the CoPilot experience is very poor.
• If CoPilot were to take over the running of my PC, I suspect that I would quickly move to Mac because at least I know that Apple is so bad at AI that they won’t try a trick like this for many years.
• What I like about my AI PC is that it delivers the same performance as its x86 equivalent but gives me double the battery life, which is what I suspect is driving virtually all of the success that Qualcomm is seeing in PCs.
• What none of the replies address is why the reaction has been so hostile, and here I suspect that the answer lies in how much better moving to an agent makes the user experience on a device.
• The desktop computer and the smartphone have long-established and high-quality user experiences with which almost all users are entirely satisfied.
• This means that moving from keyboard, mouse and touch to an agentic system will result in a small increase in utility, meaning that there is little incentive to make the switch.
• If, however, there are device categories where the user experience is either undefined or sub-standard, then things can start to become interesting.
• For example, the vehicle where RFM Research has long argued that the touch-based icon grid that is now standard offers a poor and potentially dangerous user experience for a user who is supposed to have his or her eyes on the road the whole time.
• Here, a decent voice experience that is made possible by an LLM has the potential to offer a very large increase in utility, creating a large incentive for users to make the switch.
• This is why RFM has long argued that voice in the automobile represents a significant opportunity for the OEMs to regain some of the ground that they have lost to Apple and Google.
• Other areas of potential are smart glasses, whose recent surge in popularity is almost entirely due to the improvement in user experience enabled by LLM-powered voice and the Metaverse more generally.
• The emerging robotics segment also has great potential for an agent-powered user experience, as voice will be the most natural way for humans to communicate with these machines.
• Hence, while RFM sees great potential in agents in some device categories, it remains sceptical about whether the use of agents will take off in others.
• This is why I suspect that Microsoft’s statement that it will evolve Windows to an agentic OS drew such ire and why Microsoft might end up quietly dropping the idea.
• This is very similar to what has happened with AI PCs, which were launched with Arm processors and originally were marketed as PCs with amazing new AI services.
• This soon switched to great performance and double the battery life of x86, as it turned out that users were not that interested in what CoPilot had to offer, which I don’t think is going to change anytime soon.
• Hence, I think that if agentic experiences are going to become widespread, they will be driven by other device categories, which are then brought over into smartphones and PCs at some point in the future.
• I still think that agents are going to have a profound impact on the digital experience, but it is going to begin in automotive, smart glasses and the Metaverse.
• Anyone who is banking on the imminent take-off of an agentic-based user experience on the PC or smartphone had better think again about their business model.