The Agentic Future is Here. Are We Ready?

Large Language Models like ChatGPT and Claude might already be writing full length novels and passing law and medical exams, but Open AI’s Sam Altman describes them as “incredibly dumb” compared to AI agents.

IBM defines an AI agent as, “a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its own workflow and utilising available tools.”

Unlike LLMs, which rely entirely on user prompts, AI agents can act without any human input.

AI agents range from simple task-specific programs to sophisticated systems that combine perception, memory, reasoning, and action capabilities.

They can exist completely online, or be housed in anything from sensors to humanoid robots.

There are multiple types of agents:

• Simple Reflex Agents act solely on immediate sensory input, following predefined rules to respond to environmental stimuli without memory or learning.
• Model-Based Reflex Agents maintain an internal model of the world to infer unobserved states, enabling them to operate in partially observable environments with more informed decision-making.
• Goal-Based Agents use planning and search algorithms to determine the best sequence of actions, consider future consequences and work toward defined goals.
• Learning Agents improve over time by learning from experience and feedback, adjusting their behaviour to optimise performance in environments where pre-programmed knowledge isn’t sufficient.
• Utility-Based Agents focus on maximising overall utility as they navigate between competing goals.
• Hierarchical Agents break complex tasks into subtasks which are then managed by lower-level agents, enabling organised control and efficient task coordination.
• Multi-Agent Systems (MAS) interact within shared environments to achieve individual or collective goals, using defined communication and coordination protocols.

AI agents hold the potential to redefine governance, making it more efficient, transparent, and proactive.

They could be used to maximise budget outcomes and increase resource management efficiency, improve data analysis, create uniquely tailored communications between governments and citizens, or completely revolutionise healthcare.

According to research conducted by Deloitte, AI-driven automation could save governments hundreds of millions of hours, representing billions in cost reductions.

In the future, we might see AI agents that anticipate potholes before they form and automatically schedule maintenance, or identify outbreaks from subtle data and prevent epidemics from occurring.

They might even be used to directly update citizens on proposed policy changes and collect real-time feedback to inform decision making.

The radical transformations that AI will enable is hard to fully comprehend, but we are already starting to see examples of the benefits that could flow from approaching technological advancements.

 

Increasing government revenue:

AI is already being used to identify law-breakers and issue fines.

In France, AI analysis of aerial imagery led to the detection of over 20,000 undeclared swimming pools, which resulted in over 10 million euros in tax revenue.

In Australia, thousands of drivers a week are paying the price for using their mobile phones behind the wheel.

In the future, agents could also be used to manage payments and disputes making fine issuance and management completely autonomous.

 

Customer service chatbots:

Chatbots have been used for many years, but as LLM sophistication increases and agentic powers are incorporated they will be able to tackle increasingly complex tasks and challenges.

The USA is using AI chatbots that act like refugees to train immigration officers.

In the United Arab Emirates, the chatbot RAMMAS has processed close to 10 million enquiries and has a customer satisfaction rate of 95%. As the technology advances, chatbots will be able to do more than answer simple queries.

They will be able to converse in multiple languages, adjust their outputs based on the unique profiles or personalities of their users, and perform real-world tasks that remove the need for any human interaction.

 

Disaster prevention and response:

AI can enhance disaster management by improving early warning systems and creating rapid response capabilities.

The United Nations has identified 27 use-cases for AI-led disaster management and prevention covering everything from cyclones to Tsunamis. Japan is using AI to bolster disaster prevention and mitigation for earthquakes, and the Xprize recently announced a competition for autonomous wild fire detection and response.

 

Healthcare and mental healthcare:

From research, to administration to diagnosis and treatment, agentic AI is set to revolutionise healthcare. It will also affect how we treat mental health challenges.

The USA is using AI to prevent veteran suicides and Korea is using AI to monitor and uplift lonely elderly citizens.

 

Policy design and scenario modelling:

In the future AI agents could be used to help policy makers brainstorm and devise new policies or communications campaigns.

For example, Stanford has recently created a virtual town where AI agents interact like real humans and Microsoft has recently launched TinyTroupe, an agentic service that facilitates persona research.

These types of ‘synthetic labs’ could serve as creative ways to test and refine new ideas before getting direct feedback from actual human constituents.

AI agents have the potential to transform society for the better, but they could also lead to chaos and disaster.

The harms caused by bugs, errors, hallucinations and cyber security threats are all amplified in an agentic world.

We’ve already witnessed self-driving cars colliding into trucks and killing their passengers, AI chatbots telling their users to ‘please die,’ and the tragic death of a 14 year old boy, who committed suicide after being seduced by a Character.ai version of Daenery’s Targaryen into joining her in the afterlife.

As AI agents are embedded into our daily lives, we run the risk of becoming so dependent on them that we lose the essential skills required for society to function – which could leave us vulnerable when systems fail.

Experts are also warning that rogue AI agents, if left unchecked and unrestrained, could escalate from being a nuisance to causing catastrophic, or even existential, harm.

When AI goes wrong, it can go spectacularly wrong. In Australia, the Robodebt fiasco saw over $700 million inappropriately recovered from hundreds and thousands of individuals, leaving them financially and emotionally devastated.

For a more alarming example we can look to the high-frequency trading algorithm that caused a flash crash which wiped $1 trillion from the U.S. stock market in minutes – a stark reminder of how fragile systems can be when AI systems are not properly understood or controlled.