Imagine logging into your computer on a dreaded Monday morning and saying, ‘plan my week.’ Instantly, your AI assistant springs into action, organising your schedule, booking meetings, and adapting to real-time changes – all seamlessly and autonomously.

This is the vision Bill Gates recently shared when he talked about the future of AI agents, where ‘there are no app stores’, only smart, autonomous assistants that can understand your needs and handle the tasks for you. It’s a future where AI agents are not just passive responders but active participants in our digital lives. Instead of asking for answers, we’re asking them to take action.

So, what separates an AI agent from a traditional digital assistant? Simply put, an assistant might retrieve information or answer queries, but an agent goes further. It interprets your intentions, creates a plan, and executes that plan autonomously.

“In essence, an AI agent isn’t built just to understand – it’s built to do.”

Here’s an example from our AI Labs team, where the agent breaks down a complex prompt into multiple task-driven steps. Each task is fully customisable, allowing for simple or complex actions as needed.

Screengrab of AI interface. At the top of the interface, the user has entered a prompt: "research generative AI use cases for the UK public sector, summarise into 5 bullet points and email to Nathan." Below the prompt, there’s a detailed plan with the following steps: Plan: "First, I will browse the web to research generative AI use cases for the UK public sector. Then, I will summarize the findings into 5 bullet points. Finally, I will draft an email to Nathan with the summarized information and send it to Nathan's email address." A list of tasks with checkboxes next to each: Research Generative AI use cases for the UK public sector – described as "browse the web." Summarize findings into bullet points – described as "create a summary from text." Draft email to Nathan with summarized information – described as "create a draft mail." Send email to Nathan – described as "send mail to user." At the bottom of the checklist, there is a black button labelled "Proceed,"

To understand how AI agents operate, let’s break down their functionality into three key steps: intention, planning, and acting.

Intention: Making sense of it all

Before an AI agent can act, it has to understand precisely what it’s being asked to do. Language, however, is often anything but precise. Small ambiguities can lead to very different interpretations, and this is where an agent’s ability to disambiguate becomes essential.

Let’s consider a few examples:

  • Lexical ambiguity: “I saw a bat fly across the field.”
    • Are we talking about a winged mammal, or a piece of sports equipment?
  • Referential ambiguity: “Joe gave the book to Bill because he liked it.”
    • Did Joe like the book, or did Bill?
  • Syntactic ambiguity: “The dog is ready to eat.”
    • Is the dog about to eat, or have our dinner plans changed?

AI agents are getting better at disambiguating these nuances, but it remains one of the most challenging aspects of developing effective, autonomous systems. Only by understanding intention accurately, an AI agent can move from vague commands to precise actions.

Planning: Charting the course

Once an AI agent understands the task, it requires more than a fixed set of instructions – it needs a strategy to achieve its goal effectively. Unlike traditional RPA systems that rely on rigid, rule-based models, today’s AI agents embody the principles of hyper-automation. They can adapt to changes and respond to unexpected events in real-time, bringing a new level of flexibility and intelligence to automated processes.

Imagine an AI agent managing the entire loan approval process in a financial institution. Beyond tracking application statuses, this agent coordinates each phase, anticipating bottlenecks and adapting in real-time to factors like regulatory requirements, client background checks, and market conditions that impact risk. Rather than following a rigid checklist, it builds contingencies and ensures compliance while expediting approvals – a dynamic shift from today’s reactive systems.

To orchestrate this, the agent integrates data from diverse sources, including credit histories, income verification systems, market data, and regulatory updates. It navigates inconsistent formats and fills data gaps to translate information into actionable insights, refining its approach to keep approvals on track and efficient.

“Managing real-time uncertainties is where today’s agents are truly tested.”

By learning from past cases, a well-designed AI agent in a loan approval process can reassign tasks, flag risks, and adjust timelines as new information arises, continually boosting efficiency.

Further, balancing compliance, speed, and client satisfaction is essential, and advanced agents use multi-objective optimisation to create flexible, context-aware plans that meet the evolving demands of the financial approval process.

After charting the course, the AI agent must now bring the plan to life through action.

Acting: Where the work happens

The final step is where the magic happens: acting. Once the AI agent has identified its goal and crafted a plan, it needs to act. Here, an agent stops being a passive helper and becomes active, capable of executing tasks across platforms and applications.

At its simplest, an AI agent can act by using APIs – interfaces that allow it to interact with other systems. For instance, if an agent is processing loan documents, it might connect to a verification system’s API to check applicant details, confirm approvals, and update the status in a central database. This ability to connect with APIs is powerful, but it’s only the beginning.

More advanced agents can autonomously navigate entire applications, moving seamlessly from one task to the next, much like a human user. Imagine an agent that logs into your email, searches through specific threads for required documents, retrieves relevant attachments, and then switches over to a compliance system to update records – coordinating across platforms to complete complex workflows efficiently.

In the most sophisticated scenarios, agents can operate across multiple digital environments simultaneously scheduling appointments, sending follow-up messages, and retrieving information from various sources in real time.

“Looking ahead, the capabilities of AI agents promise to revolutionise not just personal assistance, but entire industries.”

The future of AI agents

Here’s where it gets exciting – AI agents are evolving to become more than just helpful assistants in our personal lives, they have the potential to transform entire organisations. Soon agents could act as smart layers over legacy systems, taking on the heavy lifting of integration, automation, and coordination, all while providing a seamless, modern experience for users.

For businesses with deep-rooted systems, this means they can innovate and adapt faster, bringing new capabilities to life without re-engineering everything behind the scenes.

By acting as dynamic connectors, AI agents bridge silos, unify data sources, and automate workflows, creating a streamlined interface across diverse platforms. They’re making it possible for organisations to enhance user experience and optimise operations without the complexities of massive transformation projects.

In Bill Gates’ vision, it’s a future where “you won’t have to use different apps for different tasks” but have agents that can take on any task with a simple command. That future feels closer than ever.

About the author 

Nathan Marlor leads the development and implementation of data and AI strategies at Version 1, driving innovation and business value. With experience at a leading Global Integrator and Thales, he leveraged ML and AI in several digital products, including solutions for capital markets, logistics optimisation, predictive maintenance, and quantum computing. Nathan has a passion for simplifying concepts, focussing on addressing real-world challenges to support businesses in harnessing data and AI for growth and for good. 

Talk to us