Context, Communication, and Agency: AI, LLMs and The Future of Collaboration

Collaboration can be broadly defined as how we accomplish things, together. As humans, we have developed natural ways of working with others towards a common goal — or at least, we do our best, to varying degrees of success.

With Large Language Models (LLMs) becoming operationalized in product, we have new collaborators in the mix, and they’re really good — they’re fast, tireless, and never make mistakes (ok, currently, they definitely makes mistakes, but as time goes on, error-checking mechanisms will emerge as standard practice). At the time of this writing, those fully fledged and capable digital collaborators have not yet met their potential, but it’s clearly right around the corner.

What does exist now, are tools that can assist in certain areas of a collaboration, amplifying human efforts. All it will take is building those blocks into a cohesive kit, and we will have something akin to a JARVIS-like agent. One that knows all about you, your preferences, the task at hand, and the ability to access all platforms needed to execute those tasks as directed. As of November 2023, we are so close!

The first piece to align on is the goal — what counts as a successful outcome. In its component parts, completing a collaboration can be examined across a few dimensions. I like to break them down to context, communication, and agency.

Components of Collaboration

Context refers to the shared understanding among all of the collaborators. Who each member of the team is, what their roles are. Understanding the nature of the task, is it a stand-alone task or part of a larger initiative? What information or tools are needed to assist the collaboration?

Communication is the exchange of information, ideas, and updates with other members in the mix that keep the collaboration moving forward (how to define “other members” I’ll get to in a minute). Information is relayed through words, numbers, images, charts, interfaces and the like.

Agency is the ability of members to act independently within the collaboration, making decisions and taking actions that align with the shared goals.

This is where things get interesting. I would define a “collaborator” as a member that operates in all three of these dimensions. There are many tools and platforms that act as facilitators of communication, or as records of context. But to truly collaborate, a player must be able to know, communicate, and act.

Examining collaboration across these dimensions will allow us to understand more about

• How to best apply current technologies to areas of collaborative work
• Determining where there may be opportunity for new tools and processes
• And ultimately, envisioning how AI and LLMs could evolve from mere tools into true collaborators

A personalized collaborator — the Travel Agent

I used to travel a lot for work, and we had a travel agent — they were great! They had my travel preferences, access to my calendar, my rental car favorites, hotel likes and dislikes, all of the rewards account numbers. Everything was tied into my company expense account, so records were kept in the way the company required.

When I had to take a trip, there were really only a couple of simple steps:

1. Fill out an online form with the very basic information — purpose of the trip, where I was going and when.
2. If there was any additional information needed, the agent would reach out to clarify.
3. I got back from them a nicely formatted list with few options for flights and hotels.
4. I picked the options I liked best, and the agent did the rest. Everything was booked, paid and confirmed. All notifications were sent, and the expenses were filed automatically.

Let’s look at this collaboration across the three dimensions.

The context is the main goal — I’m going on a business trip, and I need to travel from here to there, and need to stay for a few days. The detailed context is my preferences, the travel dates, the budget, and any other factors that are important to consider (maybe the entire team is headed to a trade show, and everyone should stay at the same hotel).

I communicated the needed details to the agent through an online form — providing the purpose of the trip, the destination, and the dates.

The travel agent, as their name implies, used their agency to

• find possible flights, cars and hotels that fit the Goals and Context.
• Gathered the options for flights, hotels, and car rentals.
• Presented those options to me in a clear and easily selectable way, along with any special considerations about those options (i.e. this flight has a connection)

I reviewed these options, communicated to the agent my choices, and the agent then completed the bookings and handled the expense filings.

I received the confirmations on my various apps (calendar, flight, hotel, car). All I needed from there was to pack my bags and get to the airport.

The digital collaborator

Artificial Intelligence (AI) and Large Language Models (LLMs) are revolutionizing the way we collaborate. Like the travel agent, they too can understand context, communicate effectively, and, at the moment in limited ways — exercise agency in the digital interface to the real world. LLMs can parse a corpus of data to understand context in relation to the stated desired outcomes. It can then communicate with you or other team members, presenting personalized options or updates. With the right integrations, it can act with agency, completing tasks, and seeking human intervention when needed.

A travel agent is a simple, customer-facing use case, but think how this could be applied to large swaths of knowledge work — project management, research, strategy — all of these are heavily collaborative activities that require inputs from several users and interaction with several platforms. If properly implemented, AI could handle huge chunks of that work, and with a full understanding of key factors in real-time.

Here is a question that keeps my mind spinning late at night — if all teams had a super-capable digital assistant, do we even need to interact with reference platforms, or can the digital collaborators do that behind the scenes? There are many platforms that exist as a place to manage collaboration. The more capable digital collaborators become, we as humans may only need a simple, contextual, evolving dashboard of a project. Approving or directing the flow of a set of tasks that are executed with as little input from us as possible.

The Future of AI in Knowledge Work

As we look to the future, it’s clear that AI and LLMs have the potential to redefine collaboration in knowledge work. They promise efficiency, personalization, and context-awareness that, properly applied, could lead to vast increases of individual and organizational capability. It’s not a free lunch — like all previous transformative innovations that have come before, this technology forces us to consider difficult questions. How do we maintain the human touch in a world of digital collaborators? What activities can we let digital collaborators do, and where do humans need to be in the loop? How much productivity increases can we hope to see? And perhaps most importantly, how do we ensure that these technologies are used responsibly, augmenting human capabilities rather than replacing them?

These questions are often raised in articles about LLMs and knowledge work. And for good reason: they are not only important questions, but I see a deep truth about us at the core of the answer. As humans, we have been collaborating since there have been more than one of us. Millenia Ago, early man shared fire with one another so they wouldn’t have to start from scratch, and so others could concentrate on different, important tasks. In 2023 A.D. we can do more with less via digital interfaces to the real world. We can do things better, faster, more completely, and more collaboratively.

We keep zooming out. Without getting too woo about it, this is the advancement of our species. We collaborate to build upon the ideas and efforts of others, as a team. We extend our reach, and enhance our touch.