A CIO’s Guide: Understanding virtual assistants, copilots, and AI agents

In this ever evolving digital landscape that the author wrote this article in, it can be hard to stay abreast of advancements in AI. Since the term 'Artificial intelligence' was coined in 1955, there have been a lot of changes but not as many in the last couple of years since the launch of ChatGPT.

With every company looking to incorporate AI in its own product offering and its employee productivity stack, it can be quite a challenge for CIOs to figure out what kind of AI offering they should integrate and for what use case. So, we at Atomicwork decided to put together a framework to help CIOs make sense of the AI landscape and the rules of thumb(s) they need to keep in mind.

Chatbots and IT support: no AI, pure HI

The first chatbot, ELIZA, preceded ITIL by at least three decades but the impact of ELIZA on IT support exists till today.

Till recently, IT teams attempted to reduce repetitive tasks through automation scripts and simple text-based decision trees. By following the predefined script and the tree, chatbots can handle common queries by following a set of programmed rules, offering limited but useful automation for repetitive support tasks. This allows IT to offer support for common use cases online, through a portal, and reduced the number of people calling them up for support.

When is it useful?

Chatbots are perfect for use cases where

• There is no reasoning, cognitive intelligence and multi level prompting are not needed.
• Keeping data safe, secured and locked down is of the utmost importance
• Dialogue flows and use cases are already well known so all the IT team has to do is wire it up (piece of cake, amirite?)
• The answers are to be served as is, without any help from a generative AI model to summarize the information and extract key points. For example, legal documents should not be summarized or reworded.
• The user is looking only for answers from documents that already exist in FAQ formats so it’s a matter of simple search-and-surface. This means, the teams also have to create documents in this precise structure for good chatbot-recall.

However, service desk chatbots are mostly available only through portals so you have to continuously direct your employees to the IT portal every time they have a question for you. You also have to be precise about the dialogue flow design else users would end up 'stuck' because the bot didn’t know how to proceed further. This means that you have to account not just for the terms a user might use (and all of its synonyms) but also, make the interactions feel conversational like they were happening with a human.

Allocating resources for development, deployment, and maintenance for such a project can also be a challenge, especially for smaller organizations. Your systems have to be properly integrated else the user will be forced to repeat everything they told the bot which is an unpleasant user experience.

From chatbots to AI assistants: Streamlining workflows and providing IT assistance

An AI Assistant leverages artificial intelligence to enhance user interactions, automate tasks, and provide efficient and personalized support across various domains and platforms. AI Assistants use NLP to understand user queries and generate human-like responses, enabling intuitive and seamless interactions. They can maintain context over multiple interactions, allowing for more coherent and relevant responses.

Enterprise AI Assistants can be:

• Domain agnostic like Copilot that helps employees perform various work tasks like email creation or summarization, meeting scheduling, search-and-surface from personal documents and so on.
• Domain-specific assistants like Atom that helps IT teams provide efficient IT support to employees. Atom has been configured to provide answers to frequently asked IT questions (so it’s an IT expert), guide employees through troubleshooting solutions, autonomously solve some issues like password resets end-to-end, and create requests on their behalf so a human agent can help them

As with most tech, AI Assistants are most effective when they are integrated with the other apps that you use so that they can leverage APIs to perform actions or pull data in other apps. For example, using Copilot to pull deal information from Salesforce so you can prepare for your meeting with a customer.

When is it useful?

• AI Assistants can be incredibly useful for IT and HR support. They can not only handle everyday queries like 'What is the leave policy?' but also more complex queries like 'What is my leave policy?' by integrating with other systems like your HRIS or your attendance software and automating multi-step workflows like not just retrieving the time-off number but also applying for time-off, securing approval and updating the status.
• AI Assistants are very much in the 'come as you are' documentation camp. You don’t have to spend a lot of time preparing your documentation for training; often enough, it’s a matter of connecting your systems, uploading your documents and just testing the Assistant. This means you don’t have to maintain and update separate data stores or spend time on specialized formats.
• Another reason that AI Assistants end up being more useful than chatbots is that they can be easily integrated into an employee’s flow of work. Instead of going to a portal every time they need help or need to look something up, an employee can just DM an AI Assistant on Slack or Teams or email and receive the assistance they need. AI Assistants bridge the support gap by taking support to the end-user.
• AI Assistants can provide incredibly personalized support. AI Assistants can tailor interactions based on user preferences, permissions, behaviors, and past interactions to provide personalized and relevant assistance. AI Assistants can also be configured to produce answers verbatim, without any generation or summarization capabilities at play - an extremely useful capability when one is working with legal documentation.

What does it take to realistically use an AI Assistant?

I firmly believe that end users who have questions and seek answers are not interested in the process of obtaining those answers. Their needs are simple: they want their questions answered immediately and accurately. The challenge for service organizations is to fulfill this need as efficiently as possible. Atomicwork recognizes this gap and has developed a product that will revolutionize the way we structure and access information, ultimately fulfilling the end user's objective. Syachfri Tjhia, Head of IT, Catalyst Education

Copilot, take the wheel: Decision making support for agents

The terms 'copilot' and 'AI assistant' are often used interchangeably, but they imply different roles and functionalities depending on the context.

An AI assistant is designed to help users by performing tasks, providing information, or facilitating processes through natural language processing and machine learning. A copilot is an advanced assistant that works alongside a user, often in more complex or specialized environments, to enhance productivity and support decision-making processes.

Some popular copilot (small c, copilot) examples are GitHub Copilot or even Atom Assist which helps IT agents by auto-linking and recognizing major incidents, summarizing long tickets easily or providing writing assistance while responding to users.

Chatbots and AI virtual assistants typically end up being more useful for end-users whereas copilots are meant to help IT agents move fast and be more efficient with their time and energy.

When is it useful?

This seems like a no brainer but we thought we’d include it anyway. Much like an AI Assistant, a copilot is most useful when it’s well integrated into the tools your agents use for their everyday tasks. Otherwise, the copilot will just lack the necessary endpoints to be effective and your agents will spend more time switching tabs to get answers than doing the work.

For example, ChatGPT can also provide writing assistance but it requires your agents to switch tabs, write a prompt with the user’s ticket history (or use a primed chat tab), copy in their reply and wait for ChatGPT’s response. Compare this to the ease of hitting 'Ask Atom' in the reply window and just choosing the assistance you’d like.

Well-integrated copilots can go beyond level 1 tasks like content creation and correction to complex decision making assistance like:

• Prioritizing tickets for you and routing them to the right agents and teams without a human in the loop,
• Suggesting the best route of action based on previous tickets
• Escalating a ticket based on its reading of a user’s sentiment
• Analyzing tickets and providing insights and recommendations for where processes can be tweaked and outcomes improved

AI copilots are termed as such because they crunch the numbers to come up with recommendations (tasks that not many humans are skilled at doing) but still need a human to take the decision.

However, if a copilot is not a native offering from the products you use, integrating a copilot with existing systems and workflows can be challenging and time-consuming. The introduction of a copilot might also disrupt established workflows and require significant training and adjustment for the user.

Our recommendation: Go with an ITSM product that has a robust copilot offering so that it can take some of the load off your agents. Building it on your own will end up 1.5xing your agents load because they’ll have copilot maintenance to take care of as well.

Meet AI. Agent AI.

Gartner, tech’s Webster, defines AI Agents as “…autonomous or semiautonomous software entities that use AI techniques to perceive, make decisions, take actions and achieve goals in their digital or physical environments”.

AI agents can operate with minimal human agent intervention, making decisions based on predefined rules, learned behaviors, or real-time data. Think of them as trusted employees that you can send out on tasks with goals, roles and outcomes clearly defined.

AI agents can even work in crews with one agent performing a task and turning it over to another agent for processing and performing its task. For example, an agent crew can research (”read” code), write and edit a knowledge base article all on its own without needing a human to come in at any point.

So, when are AI (IT) agents useful?

• AI Agents can analyze tickets in a helpdesk and suggest (and even write) documentation based on knowledge and training gaps that it observes
• AI Agents can look for workflow opportunities based on your human agents’ activities and suggest workflows to admins
• AI Agents can step in to troubleshoot common and simple IT issues, if a human agent is not immediately available. For example, if an employee is having trouble with their VPN setup, an AI agent can “look” it up for the employee autonomously and suggest solutions based on its findings.

AI Agents have the potential to be the most useful tool in an IT team’s arsenal. AI Agents have such a high level of autonomy that IT teams no longer have to worry about helpdesk staffing predictions and can focus on moving up the Maslow hierarchy of IT needs. IT teams spend most of their time involved in tasks that do need human decision making or intervention so AI Agents can go a long way in alleviating some of that frustration and freeing up their time for more creative work.

Now that you're familiar with all the latest AI tech terms, we’d recommend you think of where and which AI is most needed in IT support with this framework.

The terms chatbot, AI assistant, AI copilot, and AI agent are often used interchangeably which is what led to the creation of this article. Often enough, when a company talks AI, their offerings tend to be a mishmash of the latter three which can lead to some confusion on the buyers’ front. The distinction, however, is quite straightforward and is purely based on autonomy and its ability to handle complex multi-turn workflows.

If you need more assistance, in choosing the right type of AI technology for your enterprise use case, reach out to us!