AI Agents and Workflows of the Future

By Dikshant Dave • 10 min read

Will AI take away our jobs? Will the way work is done see a large disruption? Way more than any other disruption has in the past?

The answer is a bit nuanced, and it would help to look at what work means today and how it has evolved over the years. Let's look at how work gets done in the modern world. (By modern, I mean after computers entered the scene) In this essay, we'll keep our focus solely on the realm of information technology (IT), (software, data etc.) leaving out the evolution of work or the role of technology or robotics in heavy engineering or mechanical industries.

Up until now, a typical workflow (in IT) in most businesses or professional outfits has involved the following components:

1. Data: This would typically include all information pertaining to a business, such as customer data, product information, inventories, process data, manuals etc.
2. Tools: By tools, I mean anything that helps you produce an output based on a given input. These could be software programs or machines, and as simple as a basic photo filter or as complex as a program producing predictive results using machine learning.
3. Connectors: These provide Interoperability and are programs like Zapier which assist in connecting two different data or business units and help them work together.
4. Managers: This role is usually carried out by humans and involves achieving the desired goal with the available resources and tools. It happens to be the most vital role in the entire process since it requires marshaling all the other units into working together towards meeting the larger business goal.

  With advancements in technology, each of these units has experienced significant innovations and enhanced sophistication.

Data

The first wave of the impact from technology was on data. With better processing power, more memory, larger storage capability, and improved user interfaces, data gathering gained a lot of sophistication. We could now store hundreds of millions of records in a single computer and retrieve them at will in a matter of seconds. This ease made businesses and institutions less conservative when asking for more information from their customers and partners.

Tools

As data became abundant, we saw technology having an equally enormous impact on tools, more specifically Software applications. More data demanded more sophisticated tools to be able to manage, handle and process it. Thus far and going forward we see software development continuing to advance by leaps and bounds.

From a business workflow standpoint, software programs can perform complex operations in split seconds. A manager can now punch in data about a customer and easily generate a profile report which can help them create a service plan tailored especially for the customer.

This growth in software engineering and development over the last couple of decades has impacted not just the application/business layer but every level, from operating systems to middleware to user interfaces.

A typical business in the present day has an average of over 50 different software services or applications, either in the form of a third-party SAAS or their own proprietary systems. Over the years we have seen newer and sharper software products arriving and challenging the incumbents. And as businesses we are accustomed to evaluating newer technology products and replacing the old ones if they show a significantly high advantage. We also see completely new software being launched in a niche vertical for an unaddressed need, seeing adoption from the players in that niche.

As for the business workflow, where these software applications play a role, they accomplish a specific task and produce a result that the unit manager then uses to progress towards the workflow goal.

Connectors

The rise in software development in general has created a distributed network of programs and data sources. Today no business can rely solely on its own programs or data. There is an increasing reliance on third parties who have established themselves as hubs in their area of expertise and provide their services via APIs. There is a complex network of interdependencies for practically any business today that uses multiple other third-party APIs to run its business operations.

Time and again new standards of interoperability do get formed and propagated, but it is hard to bring everyone on the same page. Connectors basically solve this problem by absorbing the complexity of different standards and interfaces. They are the components of the workflow which enable the “Interoperability” of various distinct third-party services, or more specifically APIs. Zapier is a good example of this - it enables a business to connect different services in its workflow and has them all working together. Connectors allow the technology teams or managers to create a daisy chain of various software programs that use the outputs of the others as their inputs and further process the information or data to produce higher-level outputs.

Managers

The controllers of any workflow are its managers. Do note that manager here is not necessarily a designation but more of a role. In a particular workflow, a software engineer could be a manager himself. A Manager’s primary responsibility is to achieve the goal for the workflow they are managing. They are entrusted with the decision-making and control of all the components of the workflow to ensure that the given goal is achieved.

To understand this role, let's take the example of a Sales Manager. Their role involves interactions with prospects who have made an inbound inquiry or the outbound leads generated by the marketing team. They basically engage in a conversation with the prospect (over emails, chats or calls), provide all the necessary information about the product (or service) and about the company from their data repository, answer questions, understand specific needs, offer a solution (by consulting with other colleagues or from past sales data), pitch the product and then if the prospect is interested and agrees, schedule a demo call with the sales director. In this workflow, the goal set for the sales manager is to qualify the prospect and convince them to agree to a demo call with their senior.

All the other components that we talked about earlier have a defined role and operate within a predictable environment of inputs and outcomes. However, the role of a manager, which is, the orchestration of all the other components - Data, resources, software, technologies, and connectors to achieve the workflow goal isn’t predictable, and more importantly involves decision-making at a business level. A Manager has the awareness of the context she is in and has the ability to handle novelty. The unpredictability of the outcome is quite high, despite her best efforts and intentions, achieving the goal may take longer than expected, yield less-than-ideal results, or possibly the goal may not be achieved at all.

In the past two decades, whenever we have spoken of technological advancements, it has most certainly meant advancements in tools and software applications (both in frontend and backend levels). This means that we implement a new tool in the workflow (or replace an older one with a newer, more advanced and more efficient alternative), which is primarily controlled by its manager. After all, the Manager is the entity that ensures that all the units of the workflow are optimized towards the achievement of the goal.

In some verticals and workflows, software applications have been advancing at a terrific pace. With the help of connectors, they cover a much larger ground, enabling the manager to be way more efficient, if not making their role entirely redundant. A good example of this is e-commerce. A medium-sized business running on Shopify can automate the entire workflow right from order booking to the shipment of the order just by using Shopify and other services available on its platform, via third-party plugins or apps. The same thing two decades ago would have needed at least a handful of managers to achieve the goal.

While the above scenario would be true in e-commerce and a few of the verticals and use cases, there are many other verticals where software applications have played a relatively more minor role, i.e. the Manager is still the controller-in-chief, and it is nearly impossible to imagine the same workflow without them. Software applications do get upgraded, often making it easier for them or for other entities to operate more efficiently but don’t make them redundant. At least not until now.

The Age of Conversational AI

As we enter this new age where Chatgpt is a household name and generative AI is starting to appear in our lives in multiple ways, the age-old question, “Will AI take away our jobs?” or an even more dystopian thought, “Will humans have no role in the future?” is again in front of us.

For the earlier disruptions caused by computers and information technology and even by the earlier generations of AI and Machine Learning, this question was eventually answered with the outcome that all these advancements made us humans significantly more efficient and productive - better managers. So what about now - Is it going to be the same as what happened earlier or is it different this time? Will AI eat humans? I am going to attempt to answer this question, primarily because the mission of my current startup, Zigment, is quite closely attached to this subject and we are keenly interested and vested in the outcome.

As we saw in the earlier sections, the development and advancement in Information technology has primarily been around the first three components of a typical workflow - Data, Tools and Connectors. A Manager's role has been steady for the most part, and even though they are becoming more efficient and resourceful, their role has evolved but stayed put. What if this changes? Is the manager’s role being replaced by a piece of software? What does it do to businesses, their workflows and ultimately - customers?

Well, this is happening already. We have stepped into the future of work. We see AI completely take over the role of a manager in the workflows of a few verticals and this AI that is taking over the role of a manager in a business’s workflow is beginning to be called an AI agent (by us and some other companies and outfits).

Rise of the AI Agent

An AI agent can also be defined by its property of replacing a human or a set of humans participating in a given workflow. Take the earlier example of a sales manager. An AI agent (pre-trained to perform this role) in this case replaces the manager and basically performs the same task, i.e. engages with the prospect, provides information and resources, understands the need, offers a solution, pitches the product/services and then schedule a call (on the calendar) with the sales director. The AI agent in this case also has the ability, just like its human counterpart, to understand the context and handle novelty.

This is not fiction, this is happening. I can say it with certainty because we have implemented the exact same use case with Zigment AI. Similar to this example, we are seeing great opportunities to implement AI agents into various use cases and workflows like travel planning, recruitment, onboarding assistance, etc. - the common theme being the manager’s role being taken over by an AI agent in accomplishing the same goal with a more or less same throughput.

It is essential to keep in mind that in the above examples, we have talked about the role of the human manager being replaced by an AI agent only from that specific workflow and not necessarily from the organization/business as a whole. The same manager could be part of many different workflows, which may or may not be disrupted by AI agents. In more complex workflows involving too many different entities and managers, AI agents could be there accomplishing sub-goals and assisting other managers in achieving larger goals.

It is only natural that a direct comparison of the AI agent would be with the human resource it replaces. But it is important that this comparison be made with the role played by the human manager rather than with the manager as a whole. The one significant aspect where humans surely win is the ability for a much broader understanding of the context, the subtle intent and the unspoken, underlying messages. But these are early days, and LLMs are getting larger and more robust. Along with that specialized LLMs for specific functions are being proposed and developed. GPT 4 has great conversational skills, while Claude is built for processing very large chunks of text.

While AI agents might still be inferior in the above-mentioned aspects, they have a definitive edge over many other aspects like available 24/7 with near instant responses. These two attributes are just impossible to have in a human team, especially when you scale. Also the fact that once programmed and trained, AI agents do not lose motivation or get tired, like their human counterparts, where fatigue is real and it is hard to keep a human manager motivated all the time. The table below shows these differences fairly well

Chatbots and Beyond

About a decade ago, we saw the emergence of Chatbots. They are usually website widgets that, as the name suggests, “chat” with users or prospects. Chatbots are an evolution from Interactive voice response (IVR), which businesses used to handle incoming phone calls for decades. As customer interaction moved from phone calls to the internet, mostly through business websites, Chatbots emerged as IVR counterparts for the web.

Chatbots are programmed similarly to IVRs—“Press 1 for English or 2 for Spanish.” They are text-based, use chat or messaging as the medium of engagement, and are configured to follow a specific path/user flow. Chatbots have been used extensively for customer support, where the user/customer/prospect leads the conversation, and the bot's role is to answer questions and provide information.

Over the last few years, we have seen Chatbots evolve significantly. Take Intercom for instance, a company providing messaging software/chatbots primarily for customer support. Intercom is integrated with the business’s data sources, such as their knowledge base, order management systems, inventories, etc., and is capable of handling much more complex queries and providing up-to-date information to the user.

However, to understand the key differences between AI Agents and Chatbots, it is crucial to see chatbots through the construct of Data-Tools-Connectors-Managers. You will notice that Chatbots haven't replaced or don’t play the role of a manager in the workflow of which they are a part. They have merely been tools to fulfill one of the tasks in the workflow which is to chat or converse with the user, mostly along the pre-scripted flow of conversation. They do not control the workflow, nor do they participate in any significant decision-making. AI Agents on the other hand are, yes, chatbots for the tasks they perform but also much more - the key difference being that they control the workflow and take ownership of the goal achievement of the larger workflow. And most importantly, they can handle novelty and the instances of context which may never have been imagined during the training. So the Chatbot comparison with the AI Agent is not entirely wrong, but it isn't the best way to understand the AI agent’s evolution and its current state.

Not to say that companies like Intercom aren’t solving a large problem - far from it. Telegram is a multi-billion dollar company and through its applications, has helped tens of thousands of companies cut down their significant human workforce, which was otherwise required to carry out the task of customer interactions, more specifically in the area of customer support. However, its value creation has been around the Tool, a chatbot and not much around becoming the Manager. It perhaps is one of the best chatbots out there but its role is restricted to being a conversational interface for customers and users (with a lot of smart features in the backend). In future, Intercom may come up with AI Agents, but that is a topic for a separate discussion.

Binary to Fuzzy

One of the defining features of an AI Agent is the ability to convert fuzzy signals to concrete actions. Let's look at the same Sales Manager example again. In their conversations with the prospect and requesting them to spare some time for a demo call with their superior, the prospect (a dog lover in this case) might agree by jokingly saying something like “Sure, but only if you promise to adopt 2 dogs from a shelter”. In this case, a human sales manager might understand the joke or the subtle nuance and know that it is a yes. The AI agent must also understand these nuances and process this conversation to go ahead and book a slot on the calendar for the demo call. LLMs have made this possible. However the AI Agent management system will need to address this complex handling of the Fuzzy-Binary signals without compromising on the flexibility of the overall system. It would involve task management and delegation between micro-agents and ensuring a constant upkeep of the overall system.

One of the abilities of LLMs is Fine-Tuning. This is basically a type of training of the AI model (LLM) with extra data laid on top of what the model is already trained on. This ability allows companies to specialize the model with their own data set, resulting in a model that understands the company’s transactions, behavior patterns, and extensive success and failure scenarios, along with the worldly information that it is already trained on.

At Zigment we are building an operating system of AI Agents (AgentsOS), which takes care of this fuzziness spectrum and task delegation along with other underlying necessities for a smooth deployment and running of the system.

Will AI Agents Eat Humans?

Will AI agents take away our Jobs? Sorry about taking a little bit longer to arrive here. The backdrop provided earlier will help me explain the answer better.

The answer is - AI agents will surely eat the roles which humans play. What this means is that Humans will evolve into playing larger roles in higher-level workflows or even managing multiple AI Agents. But a lot of current roles are going to be eaten by AI agents. One may be inclined to think that this is not too different from the earlier disruptions where machines or software took away roles played by humans. Before spreadsheet software, there were thousands of human employees punching away numbers on paper in most financial organizations, remember? However, these roles were mainly unitary tasks and not necessarily those of a manager. Managers managing the operations continued to survive (and evolve) even as the tools took away many jobs. With AI Agents we see that the role of managers, which up until now, was pretty safe, is starting to be challenged.

So which industries or use cases do we see AI Agents having the maximum impact on (or none at all)? Many of them, but not all.

Some of the heavily transactional verticals like core banking, which involve almost zero fuzziness have already evolved through software applications and such tools. Today you no longer have to go to a bank and interact with a bank teller for money transfers or deposits. All of that can be done from a banking app on your cell phone. The same would be true for an e-commerce store as well. On the opposite end of this spectrum are the workflows which are extraordinarily fuzzy and rely heavily on human interactions, extending beyond conversations. Take for example, used-car sales, where pitching to a prospect, inviting them to the showroom and scheduling a test drive can all be automated with an AI agent. However, parts of the same transaction that involve accompanying the customer on a test drive, jointly inspecting the car, negotiating prices, etc. are extraordinarily fuzzy and may not get addressed by AI Agents in the near future. In our Sales Manager example, scheduling a demo is one part of that business transaction, the other part would be to actually impress the prospect in the demo and subsequent calls to win the business finally. These may be better off with human managers for now. Most business transactions would involve multiple workflows to align together in sync, to achieve the ultimate business goal.

From the businesses’ standpoint, if the outcome of the function is too large in value, then there may not be a significant pressure to replace the human manager from the mix, i.e. the business outcome would be able to justify the costs and effort involved in having a human manager. And if the outcome is too small in value, then it might most likely get solved with tools and software applications. For everything in between, where a business wishes to have a human manager in the workflow but can’t justify the cost of having one, can now be fulfilled with AI Agents. The vast expanse of business workflows between the two ends of this spectrum showcases the current opportunity area for AI agents, from selling Insurance to booking travel itineraries to hiring - and many more.

We are in the early days of the AI age and a lot of the basic infrastructure is still just getting implemented. Coupled with rapid advancements in AI and LLMs, we are about to see massive disruptions in the way work is done. AI agents of tomorrow will look very different from what they are today, but even the ones of today allow businesses to unlock value that was never seen before.

About Zigment

Zigment is an AI-enabled lead nurturing and conversational sales platform. We help businesses improve their sales conversion by directly engaging and nurturing every lead individually to help customers make better buying decisions. Zigment orchestrates a business’s entire sales workflow with its AI agents, who help, qualify, pitch, follow up, and convert leads 24/7.

Some of the verticals that we have addressed with our technology are — Healthcare, BFSI, Automotive, Home Services, and more. If your business sells products or services which require consultative sales, i.e. any kind of consultation between the prospect and your sales team, then it would be worth considering AI Agent implementation in your sales funnel.

We would love to discuss opportunities to show you how our AI agents can help accelerate your business.