Will 2025 be the year of agentic AI? A good time to innovate responsibly

As more people are overcoming their fears and embracing the use of AI, I’ve noticed a marked increase in interest in ‘Agentic AI.’ Agents are software:

• backed by some sort of ‘intelligence
• which has a degree of autonomy
• is usually coupled with some kind of ‘sensing’ (which may be simply querying the web)
• and typically collaborates with other agents and people to achieve its objectives

Whilst the idea of agents is nothing new, the rapid increase in capabilities of the latest deep learning models – GPT-4o, Claude, Llama 3 etc. mean that what your agents can do has significantly improved.

Right now I’d say that people’s interest in ‘agents’ comes from two main motivations.

1) You want AI to address a particular task

This is being reused in lots of places (we’re seeing ‘Sales AI Agents’ , ‘Personal Assistants’ , ‘Digital Worker Agents’ etc.). I’ve started seeing this being referred to as ‘vertical agents’ because they manage a specific end-to-end process. We recently completed a project which built customer handling agents, which massively improved our client’s ability to service customer requests. This is a great example of a vertical agent – it’s focused on a specific task but being used for a variety of organisations and brands.

2) You want to make it easier to build software

Wouldn’t it be nice if instead of having to write code you could create a virtual group of people to do some things for you? Maybe collect a bunch of information, analyse it and write a report on it? Or analyse your company’s market and create a marketing strategy for it?

In an agentic framework, you specify a number of agents with different roles and objectives and ask them to work together to solve a task for you. In the ideal world you would not need to create any code – you’d specify it all in natural language (although at the moment, you definitely need some development (python) skills for most agentic frameworks.)

There are a number of frameworks available for this like Microsoft’s AutoGen or Crewai. Most of these are still maturing, and are subject to change. In my experience they are great for working with textual data – finding it, analysing and creating reports from it. They don’t yet have the ability to make things happen (like buying something or posting an advert). These need to be coded. Maybe these will start to appear later on in the year.

There is clearly huge potential for agents – and it is a rapidly changing area. In these sorts of situations we recommend starting with a quick proof-of-concept to build some use-cases you think are valuable, and get a feel for what the technology will do for you.