agentic AI: the next big next big thing
When ChatGPT launched in 2022 generative AI’s capabilities caught everyone off-guard. Advances that seemed years off were suddenly either readily available or well within range. Since then, a glut of gen AI tools, like Claude, Gemini, and Perplexity, have filled the market, column inches, and nervous boardroom discussions everywhere.
The thing is, generative AI was never the goal. People want AI that can walk the walk as well as talk the talk. Gen AI can synthesise knowledge on any topic, write copy, summarise documents, and even draw pictures of Harambe and Einstein hanging out in biker gear. That’s (mostly) genuine business value – and there are more specialised applications making huge impacts too.
Gen AI is just not that useful for most organisations – yet.
Like most people, I find myself amazed by what it can do but often baffled by what it can’t. Do all breakthroughs share the same mix of world-ending doom and adorable goofiness as generative AI? The ROI for most organisations will come when something can reliably take action in the real world, complete tasks autonomously, and make decisions. The latest releases, such as Open AI Operator and Gemini 2.0, are the first general purpose tools tentatively stepping in this direction. They mark the coming of the next big next big thing – agentic AI.
So far so sufficiently sci-fi. But, as I’m sure you’re wondering, what is agentic AI?
What is agentic AI?
AI is like raw brainpower; it needs to be given a job to be of use. Current AI tools are good at doing specific isolated tasks, like suggesting a delicious moussaka recipe for dinner. Agentic AI, however, can complete whole workflows independently in much the same way a person would. You could ask it to find a moussaka recipe on BBC Good Food then add all the ingredients to your online shopping basket. It then browses the internet, clicks buttons, scrolls, finds the recipe, heads to your online grocer, searches the items, decides which products are best, adds them to your cart and only comes back to you to confirm everything before purchase.
That’s a real example of what Open AI operator can do today. And, okay, unless you’re opening a Greek restaurant it might not generate that much revenue yet, but it clearly has huge potential. Where will we be by the end the year?
Giving a ‘digital body’ to AI that can go off and take actions – essentially automating processes on the fly – will have real impacts on businesses. Generative AI is like GPS, it can synthesise vast amounts of data to create an optimised route that you have to drive. Agentic AI is like a self-driving car. A self-driving car for your finance function, your warehouse management operations, your, um, transport and logistics department. That’s when the real transformation arrives.
We’re still a way off prompting an LLM to ‘implement transformational AI across my entire organisation’ and then taking an early lunch. In the meantime, we’ll have to do it ourselves.
By Alex Devine