Following decades of an office landscape that has largely iterated on reliable IT services built on the same hardware and software foundations that have powered the business world for decades, the biggest company in business IT has set out a bold, controversial view of the future.
In a video about the vision for Microsoft Windows in 2030, Corporate VP of Enterprise and OS Security David Weston revealed an ambitious plan to develop what was described as a more natural way of using a computer.
Whilst one aspect of it is replacing the typical keyboard and mouse with a voice-driven interface, another substantial aspect of this future is having software tools either augmented or outright replaced with agent-based systems, where a user would just ask an AI system to complete a task for them.
As with the rest of the 2030 Vision pitch, it has been controversial, but to understand why, it is important to understand what software agents are, how Microsoft and IT have been here before and whether this time is actually different or another attempt at a failure.
What Is A Software Agent?
A software agent, sometimes known as an AI agent, an agent or just a “bot”, is an autonomous application that is either integrated into other software suites or runs independently but provides help and assistance somewhat autonomously.
They can vary significantly in scope from acting as an enhanced autocompletion system, search engine or priority email inbox, to being semi-autonomous context-sensitive tools that can detect tasks and offer assistance, or respond to natural questions with relevant software commands.
The idea is that it is a more natural way of interacting with a computer, something that helps to remove the layers of abstraction that come from using a command line, a keyboard and mouse, as well as reducing the amount of time needed to teach IT skills.
Bob, Paperclips And Microsoft’s Agent Dream
Microsoft did not invent the software agent; even discounting science fiction depictions of the concept, Apple conceived of the Knowledge Navigator in 1987, which was a prototype of a software agent that was remarkably similar to systems such as Siri, Cortana and Google Assistant.
However, as early as the 1990s, Microsoft was attempting to streamline and demystify the personal computer, something that would ultimately lead to Windows 95.
At the same time as this, however, the company wanted to go even further with simplifying the computer. To that end, they developed Microsoft Bob, a virtual assistant and streamlined user interface that replaced the desktop with a digital cartoon house.
Bob was a nigh-total failure, selling just 30,000 copies and being regarded as one of the company’s biggest missteps of the 1990s.
However, the concept would be revived with Windows 98 and Office 97 with the Microsoft Agent, a software agent that would always appear on screen as a cartoon avatar that would offer context-sensitive help and assistance.
The most famous of these was Clippy, an animated cartoon paperclip that would be a constant presence for people using early versions of Microsoft Office.
Whilst it did help brand new users through its context-sensitive assistance, pointing them in the direction of templates, help files and task wizards, it was typically seen as overly intrusive and more likely to frustrate than to assist.
However, there was also a much bigger issue that could become a significant problem for businesses aiming to adapt AI agents into their workflow in the future.
BonziBuddy – A Cautionary Tale
Whilst not strictly a productivity tool or piece of business software, BonziBuddy had a similar pitch to the types of software agents that Microsoft are increasingly advocating for today, although much of its functionality was achieved through somewhat cruder means.
Marketed as the ultimate online companion, BonziBuddy used the same technology as Microsoft Agent and claimed to “learn” from the users, make using the internet safer, and provide a vaguely natural interface similar to a modern chatbot where users would type questions or requests and BonziBuddy would respond using text-to-speak.
What made the software concerning and echoes issues with modern AI systems is how much data it collected from users in somewhat opaque and underhanded ways.
It was described at the time as spyware, although it collects less data than Copilot does, and was the subject of multiple lawsuits due to its unethical data harvesting activities.
Much like how no office computer at the time should have used BonziBuddy, the security concerns of using modern AI agents should be something an IT team carefully audits before allowing their use around systems that contain sensitive data.