The Future of Business with AI

Artificial Intelligence (AI) has been, and continues to be, one of the most exciting topics in recent times. The potential for AI to accelerate business is remarkable and many businesses are already realizing the effects of these accelerations. What is both exciting and formidable about these accelerations is that we don’t yet know the maximum rate of acceleration that will be facilitated by AI – and we don’t know how long these accelerations will continue for.

It can be difficult to imagine the picture of what AI and business altogether will look like in the coming years. Looking back just four and a half years, DISTek, and many of our clients, were using Skype to communicate on a daily basis. Skype now sounds as antiquated as Internet Explorer or AOL Instant Messenger in the era of Microsoft Teams, and it’s difficult to imagine carrying out the modern day-to-day tasks of any business on a platform as primitive as Skype.

I don’t think many people anticipated the transformational effects that the then relatively unassuming Microsoft Teams would have on the business environment, and yet the modern environment is virtually unrecognizable. The expectations wagered for the imminent transformation from AI cast a shadow on the meager expectations of change that Microsoft Teams brought about. Virtually all major news outlets and tech companies are buzzing about AI and the effects it will have on the future of business.

Some amount of skepticism about these claims would be justified. It wouldn’t be the first time the potential of a new technology was overhyped by media outlets and popular culture alike. Indeed, that may again be the case with some claims about AI. However, there are already many remarkable capabilities of AI which are being realized by businesses today.

Imagine how many questions your administrative staff are asked each month. Now think about how many of those questions have answers buried somewhere within the impenetrable pile of documentation on your organization’s public website or private SharePoint site. With recent developments in AI, this documentation can be surfaced through natural language in a chat interface with an integrated AI assistant. An AI assistant of this sort can (theoretically) answer employee questions 500 times more cost-effectively than a human counterpart can[1]. Very basic assistants of this sort can be developed rapidly (on the order of minutes and hours) and be sustained for less than a few hundred dollars per month with modern tools.

Albeit, many of these modern AI assistants are far from perfect, and their pitfalls can be significant. They tend to excel in tasks like brainstorming, outlining, and data summarization, but they may fail to understand acronyms or semantics correctly. They may also find answers to questions in documents that are not good sources for that information and they might be flat-out incorrect. Based on internal testing, using public and private data, without a data architecture designed specifically for AI, a minimal configuration, and a relatively immature technology underpinning it, a basic modern AI assistant can correctly answer between 50-70% of basic employee questions, like “What are the 2024 holidays?” or “Does DISTek reimburse travel expenses?”[2]. In the 30-50% of cases where the Assistant is incorrect, it’s answers can quite convincingly steer users in the wrong direction. This ratio of correct to incorrect answers does not achieve a standard of reliability that most businesses would be willing to rely on, but some excitement may return when looking at the future possibilities for this technology.

Through dabbling with the AI-based technologies that have been published by major tech-companies so far, it’s possible to get a sense for where these companies are heading and to predict some of the capabilities of AI that are just around the corner. Some of the currently half-baked features will be cut and others will be developed into our everyday workflow and become essential to it.

Microsoft has been relatively public in their vision for AI by marketing both the current and future features of their AI assistant, Microsoft Copilot. It’s clear to see that Microsoft’s vision for AI extends far beyond the simple data summarization skills that it takes to answer employee questions. Beyond the data in SharePoint, Microsoft, with their Microsoft 365 architecture, have erected the infrastructure for customized Copilots to securely access and manage emails, Teams channels and messages, org-structures, meetings, calendars, to-do tasks, spreadsheets, notifications, notes in OneNote, and Windows itself.

Now, let’s pump the brakes just a little. Not all these features are fully developed and the underlying model which facilitates the connectivity between these endpoints leaves more to be desired (recall the limitations with acronyms and semantics). However, let’s also note that the technology is likely to improve. To borrow words from Marquese Brownley, “This is the worst this technology will ever be.” And as far as I can tell, Microsoft isn’t simply pontificating about the virtues of AI. Microsoft has put forth a credible vision, backed up by technological architecture, expertise, and partnerships to make it happen. Not only does the Microsoft 365 architecture appear to meet the theoretical technological requirements to achieve Microsoft’s vision for AI, it also makes it secure for businesses to approach confidently.

In this way, it might be fair to say Microsoft 365 is the trojan horse of AI. Consider that Microsoft 365 has a market share of about 46% and Windows has a market share of about 72%[3][4]. With this massive surface area, Microsoft’s Copilot is poised to accelerate the operations of a massive portion of businesses across the globe.

Now let’s imagine how other major players fit into the picture. Google, Open-AI, Meta, Atlassian, Github, IBM, Amazon, Apple, Salesforce, Adobe, Oracle, and countless others all have AI powered assistants that can handle various tasks within the context of their own individual environments. Each one of these provides exciting capabilities and efficiencies in their own right, but what does the future hold for these tools? Will one AI rise to the top and dominate the rest? I think there is a more likely and more desirable outcome.  

Out of all the AI tools I outlined in the previous paragraph, they all have a striking similarity that may allow them to collaborate and complement one another – language. A part of what makes each of these tools so powerful is their ability to understand the semantics of natural language and translate them into actions. Beyond keyword indexes of the past, these tools enable the use of semantic indexes. Unlike traditional keyword indexes that search based on similar, pre-programmed words, semantic indexes provide a map for the context and meaning behind words, allowing AI to interpret queries more like a human would. With semantic indexes, intents can be communicated in a practically infinite number of ways meaning that the requirements for the assistant to enact that intent are softened. The chat interface of these AI assistants can be thought of as a “soft-API” for the action-domain each model operates in.

Imagine asking a group of five of your friends if anyone has a spare phone charger. Three of them check their pockets and bags and come up empty handed, one of them finds a charging brick in their backpack and another has a charging cord in their pocket. Perfect! Your friend with the charging brick and your friend with the charging cord both agree to lend you the devices so that you can charge your phone. The two of them pass the charging brick and cord to another friend who is sitting near you, who hands both the charging cord and brick off to you. Everyone in the room is satisfied that you can charge your phone.

Now let’s imagine a couple of similar examples with AI.

You may ask one of your AI assistants to turn on the lights in your home. If it’s the AI assistant responsible for your lights, then great, you’re done! If it’s not the assistant responsible for your lights, the assistant may simply not turn on your lights and you’ll be left to manually invoke another assistant or turn on the lights yourself. This is essentially the situation we live in today.

Now, let’s imagine all your AI assistants are “friends” with one another, they talk with each other on a regular basis, understand one-another’s strengths and weaknesses, and have a good relationship. Since both models communicate in natural language, they can form a network of “friends” that can turn on your lights – even if you ask an AI assistant that is individually incapable of turning on your lights. Altogether, a network of AI assistants forms a vastly superior system to any individual AI assistant. I think our future with AI is destined to develop along this path of AI Networks.

Let’s step back for a moment and imagine the implications of this. Consider the development of civilization and the effects of written language, the telephone, the internet, and even Microsoft Teams on humankind’s ability to communicate, form networks, and ultimately become more productive. Soft APIs, supported by large transformer models (LTMs), specifically Large Language Models (LLMs) and Large Action Models (LAMs), may become the next major steppingstone in the civilization of software networks.

To re-iterate an important fact, these models are not perfect and natural language is not a perfectly robust way to communicate either. Anyone who has played the game “telephone” knows how the meaning of words and phrases can change as they travel through a network. I don’t think soft APIs or LLMs are an exception to this, but that does not undermine the utility of communicating in that fashion altogether. AI networks seem relatively certain to have a major role in our future, and the implications I can deduct from that are awesome to think about.

From where we stand, it is difficult, if not impossible, to know the extent of how AI will change the landscape of business and general productivity. I think we are in for an exciting ride! On change, Alan Watts said “The only way to make sense out of change is to plunge into it, move with it, and join the dance.” I hope everyone brought their dancing shoes!

[1] Considering licenses starting from $200/month with a volume of 25,000 messages per month.

[2] Based on the results of internal testing at DISTek. Efficiencies may vary at other organizations.

[3] Office 365 company usage by country 2024 | Statista

[4] Desktop operating system market share 2013-2024 | Statista