AI Agents are already here – and will be checking digital adverts on your behalf?

Terms like “AI Agents” and “Agentic automation” have been flying around a lot recently, with all sorts of predictions being made about their imminent importance:

• “By 2028, at least 15% of day-to-day work decisions will be made autonomously through agentic AI, up from zero percent in 2024.” (Gartner)
• 71% of organisations agree that AI agents will help drive higher levels of automation in their workflows (Capgemini Research Institute, Generative AI executive survey, May–June 2024)
• “Think of agents as the apps of the AI era. Just as we use different apps for various tasks, agents will begin to transform every business process, revolutionizing the way we work and manage our organizations.” (Charles Lamanna, Corporate Vice President, Business and Industry Copilot at Microsoft)
• “We’re in the midst of a technological shift: the transition from generative AI to agentic AI” (Paul Drews, Managing Partner, Salesforce Ventures)
• “By 2025, 25% of enterprises using GenAI will have started using these intelligent assistants, marking a fundamental shift in ‘who’ we work with and how we work.” (China Widener, Vice Chair and US Technology, Media and Telecommunications Industry Leader, Deloitte)

Now, if you believe at least some of the above to be true, you may entertain the vision shared recently by Perplexity AI’s founder that agents could even replace human attention as the target of advertising…?

What we can be certain of is that major players like Microsoft are putting serious resources into taking GenAI forwards by giving it the ability to take action on our behalf. Having an intelligent personal assistant is not a new idea, as before we encountered ChatGPT in 2022 there was Siri in 2011, Dragon Dictate in 1990, and even Bell Lab’s Audrey in 1952. But an Agent is more than just language understanding, as it needs to exhibit the following characteristics:

• Autonomy: Agents independently perceive their environment, make decisions, and take actions without relying on external instructions.
• Perception: Agents are equipped with sensory capabilities that allow them to gather information about their environment through the use of sensors.
• Decision-making: Agents make decisions based on perceived information, selecting appropriate actions to achieve their goals.
• Action: Agents perform actions that alter the state of their environment.

As the above suggests, Agents can exist in the physical world as well as the digital one with a thermostat being a common example along with self-driving cars. What most people are getting excited about though is the ability to start up a chat in Teams and ask in natural language for travel options, or to reschedule a meeting, or an analysis of the financial performance for product X. Basically, this is the ability to offload all sorts of tasks to a truly intelligent assistant who, with your input (and approval), can act like your own dedicated team of interns and analysts whenever you need them.

Clearly, having LLMs making all sorts of advances is really helpful when it comes to understanding user requests and communicating with other parties, but there’s a lot more that needs to fall in place before you can expect an Agent to handle booking your travel on your preferred (public) website. It hence makes sense that the first wave of Agents (Copilots) coming to market are inside line of business apps like Dynamics, Workday and ServiceNow as they come with the integrations and process knowledge needed to be useful.

This raises all sorts of questions (to be covered in the next post), such as:

• Security? Trust?! Just how much autonomy do we want an agent to have when working on our behalf?
• What happens to user experience when everything is potentially reduced to a chat message? Will we need to work with lots of different agents for different needs?
• How do we make sure that agents are interoperable as well as safe? Will there be agent stores in the same way that today we have app stores?
• What happens to line of business applications? Do they just become databases?!

We’re being told that 2025 is the year that AI Agents will become more mainstream, and I for one want to be involved as they start to change how we and our customers work.