The evolving capabilities of AI powered by advanced large language models are reshaping the way we work. These capabilities are bringing us closer to a future where agentic systems that are autonomous are not mere fiction but a reality. AI agents when grounded on our organisational context and processes not only can assist but also act independently where applicable. Imagine our workday if routine tasks and manual handling of work related decisions didn’t take up hours of our time and we could hand these to be completed to our own personal ai agent.

Improvements in scalable computing infrastructure have driven exponential growth in the compute power available for machine learning, making it feasible to train sophisticated models at an unprecedented scale. A sophisticated AI model (such as OpenAI GPT-4 series or Meta’s LLaMA) can generate text, analyse and describe images, translate languages with context, aim to solve problems, and even reason over given inputs, all by understanding and processing multimodal inputs.

As AI agents continue to develop with capabilities like autonomous decision-making and task prioritisation, they open doors to a future where agents are a trusted partner to a business rather than just a tool. They’ll evolve to manage more aspects of a business autonomously, yet always be grounded in principles that align with business goals and human needs. AI agents as they evolve has the potential for a future where digital technology truly works for us, creating a seamless partnership between humans and machines in conducting business.

So, what then are the desiderata* for an agent?

Adaptability and Learning: The ability for the agent to adapt to new situations, learn from feedback, and improve its decision-making over time is crucial. This might involve reinforcement learning or continuous fine-tuning based on new data inputs.

Multimodal Understanding: Given the complexity of real-world tasks, the agent should have the ability to process and understand multiple types of input—such as text, images, speech, and even video. This capability ensures it can engage with a variety of tasks in diverse environments​

Autonomy with Accountability: While the agent should act autonomously, it must also have mechanisms for accountability. This includes traceability of decisions, transparent actions, and the ability to explain its reasoning in understandable terms, especially for complex tasks or when interacting with human stakeholders.

Collaboration and Integration: The agent must be able to collaborate effectively with human operators and other AI (agentic) systems. This requires interoperability, the ability to understand human intent, and seamless integration into existing workflows​.

Robustness and Reliability: The agent should perform reliably under a wide range of conditions, with resilience to failures and unexpected inputs. This might involve redundancy in decision-making processes and the ability to function even with partial information or in uncertain environments.

Scalability and Efficiency: The agentic system should be able to scale both in terms of performance and the number of tasks it can handle, while remaining computationally efficient, particularly in resource-constrained environments​.

Safety and Ethical Reasoning: Ensuring that the agent behaves safely and ethically is monitored. This includes preventing harmful actions, ensuring fairness, reducing bias, and being aligned with human values and regulatory standards. (for the most part this is table stakes for most businesses).

These desiderata reflect a balance between technical capabilities (like processing power and learning ability) and ethical considerations (such as safety and fairness). Together, they set the foundation for agentic systems that can operate autonomously in the right settings as required.

*The term “desiderata” refers to things that are considered highly desirable or necessary. It comes from the Latin word desideratum, meaning “something desired” or “something wanted.” In a broader sense, desiderata are the qualities or criteria that are sought after in a specific context, such as the key attributes for an ideal system, product, or entity.