
The two best known AI agents that have been released by major developers are Operator by OpenAI and Manus by a Chinese startup called Butterfly Effect. Those with an OpenAI Pro account (cost: $200/month) can trial Operator, while Manus has recently become available to the general public. Platforms such as ServiceNow and Salesforce are also offering agentic AI solutions. These prototypes are not yet at the ’buy a flight’ stage. Instead, they are targeting use cases such as customer support, scheduling and procurement.
There is, however, a general belief that agentic AI will become increasingly prevalent in the coming months given the companies experimenting in the area. Adept and Cognosys, are developing agentic AI to operate software applications for example, reading an online recipe and then feeding the ingredients into your shopping list. Meanwhile, CrewAI are developing a framework for systems where multiple AI agents with different roles and responsibilities will be able to collaborate.
There is, however, a concern that much that goes by the name of agentic will be hype, as companies claim that their Large Language Model (LLM) chatbots are in fact AI agents, despite having no autonomy to execute tasks.
Autonomous AI is already creeping into real workflows. Mercedes has incorporated AI agents into its in-vehicle conversation assistance, while Bayer is using them to predict flu trends. While the world was rocked by the introduction of Generative AI in 2022 with the launch of ChatGPT, the next evolutionary stage to agentic AI is underway.
As with Generative AI, failing to prepare (or simply banning its use) won’t prevent agentic AI from entering an organisation. It just increases the risk of unmanaged and ungoverned adoption. At the same time, don’t be led astray that this is a technical concern. It’s not. It’s about strategy, culture, and operations.
The AI that we have become used to at work and at home has generally been assistive. It can assist with writing an email, with finding how to use the contents of your fridge for dinner, or with summarising large quantities of information.
Agentic AI presents a fundamental shift from these assistive AIs in that it introduces autonomous AI. On receiving a task, agentic AI will plan, act and respond to whichever tool it interacts with. This enables the agent to improve and streamline processes where possible.
To illustrate the difference, while an assistive AI will summarise a meeting, agentic AI will schedule the meeting, manage clashes, create the agenda and minutes, and take action on all items. Imagine an AI that not only schedules meetings but books travel, secures catering, and sends follow-up tasks without human intervention.
A common mistake is to see AI as a technical issue for organisations. AI is, after all, new technology. However, as we’ve seen through adoption of Generative AI, technology operates in a human context. The problems that arise are therefore human as well as technical. Take, for example, the use of shadow AI in which an organisation may ban the use of Generative AI in its policies, but its employees continue to use AI on their phones. This way the policy lacks force, and the organisation risks significant loss of Intellectual Property.
Strategy
What is your organisation’s strategy, and how does agentic AI play into achieving the strategy? Implementing agentic AI without a clear purpose or vision as to how it plays into organisational strategy will at best lead to sub-par performance, at worst seem to be an irrelevance.
Compliance, governance and ethics
Most organisations have a set of explicit values, and some even have values specifically for AI. To be effective, these need to be embedded throughout the technology stack and especially into agentic AI. Does your organisation have governance frameworks and structures in place to oversee the benefits and risks that come with agentic AI?
Technology and infrastructure
In order to function, agentic needs to interact with existing technology and infrastructure. What does your technology stack look like and how might agentic AI fuse most effectively with it? An agent will not be able to manage systems that are not interoperable now.
Data
Does your organisation have the quantity and quality of data to implement agentic AI effectively? Many have rushed in to adopting Generative AI without the core data necessary to produce anything more than disappointing, generic results. For agentic AI to function at its best, it needs to be built around reliable, good quality data.
Expertise
What’s the level of AI literacy within your organisation? This extends beyond developers or IT enthusiasts to every worker, most of whom will be interacting with AI agents in the next few years. While it doesn’t affect many in the UK, the EU’s AI Act mandates AI literacy programmes for employees of all organisations using AI. While this isn’t currently a legal requirement in the UK, it’s a sensible direction of travel, and one which forward-looking organisations will benefit from most by embracing now.
Common traps to avoid
As with any new technology, there are many challenges lying in wait for early adopters, and probably as many for those who wait and risk falling behind. At the moment, there’s considerable hype around agentic AI, and vendors could be tempted to dress up a Generative AI chatbot as an AI agent. Look for the degree of autonomy in an agent to evaluate it. At the same time, do you need an agent? If a chatbot is sufficient for your purposes, then don’t buy into the hype and stick with what you need.
Secondly, Generative (and agentic) AI displays emergent qualities over time when run at scale. There are a number of governance frameworks available, from ISO42001 to the National Institute for Standards in Technology’s Risk Management Framework and of course the EU’s AI Act. Using governance frameworks to establish a robust oversight mechanism for all AI, but especially agentic AI, will help to avoid ethical, reputational and compliance risks.
Lastly, customers and employees don’t want to be trapped in AI loops in which they are sent from pillar to post without finding what they want. We’ve all experienced frustrating chatbot loops that refuse to hand off to a human. Imagine that happening with payroll errors or contract disputes. It’s essential to look for and listen to user experience through trials and roll out at scale. AI agents can be revolutionary for organisations, but it is helpful to be on the right side of the revolution.
Leaders need to grasp the potential of agentic AI as benefit and risk to their organisation. They should consider how agentic AI can boost their organisational strategy to identify where best to invest in the technology. Run a risk and opportunity audit to find pilot use cases to trial the technology. To be an early mover in this market, it’s crucial to start small and learn fast. Don’t wait for maturity.
At the same time, plan for maturity from the beginning. Design cross-functional oversight structures and governance models that can grow with usage and complexity. These should draw from across the organisation, including IT, operations, legal, HR, procurement and others. Build on existing governance frameworks and organisational policies to embed readiness for AI agents. Consider the organisational culture and determine how best to ready employees for the changes coming, while reassuring them that their jobs are not at risk.
Originally posted here