The world may still be adjusting to generative AI and large language models (LLMs), but as Moore’s Law dictates, tech capabilities double roughly ever two years and, right on schedule, it’s now time to embrace agentic AI.
Whereas generative AI is great at following prompts to generate text, audio, imagery and other forms of content by learning from what already exists, this branch of AI is capable of creating and executing its own commands.
This shift from content creation to decision and execution machine is huge in many respects, but it also has the potential to supercharge CX.
What can agentic AI do?
Agentic AI creates its own to do list. It can understand a task, map and execute the various steps required to complete it, then feedback on the progress and results.
It does this with minimal human intervention, it can devise and employ novel solutions, and it can continually learn to improve its performance. This enhanced adaptability means it can operate dynamically in response to what is happening in its environment, and it can be deployed across applications. Furthermore, it does all this from natural language prompts, rather than code.
Because agentic AI can make its own decisions to find the route to a solution, it requires strong ethical considerations and oversight. And as we saw with generative AI, some organizations may learn this lesson the hard way.
What’s the difference between generative AI and agentic AI?
Whether it’s writing poetry or creating images and audio, generative AI’s capabilities have taken the world by storm since the general release of ChatGPT in 2022, and now agentic AI is on track to do the same. However, there are some key differences between the two technologies.
One of the core functionalities of generative AI is that it works from prompts – and the better the prompt, the better the response it generates.
This has seen generative AI become indispensable for a range of tasks, from re-writing ideas into comprehensive texts, to generating images and even adverts in order to hyper-personalized creatives to small sub-groups or even individual customers.
Generative AI has even proven its worth as a search engine. But it has limitations. Generative AI is not designed to perform continuous tasks – what it is taught is what it knows. This means it draws on the data accessible to it, as instructed by a human and as mentioned, guardrails are essential. When it comes to public models, such as the most accessible version of ChatGPT, these limitations mean responses are scraped from the internet as it was three years ago.
Agentic AI is autonomous, and as IBM explains it, this means it can complete continuous and complex tasks involving complex workflows, data-driven decision-making and action-taking, theoretically with minimal human intervention.
Agentic AI is also less of a regurgitation machine, because it can think of and complete various processes to conduct a task or draw a conclusion. Furthermore, it does this using natural language rather than code – and this is key to its widespread adoption.
In short, generative AI is a large language model (LLM) skilled in teaching itself how to use language in a human way, whereas agentic AI is focused on selecting its own actions. Or to put it another way, generative AI generates content, agentic AI has agency.
Despite the clear differences in functionality, agentic AI would not be possible without generative AI. Before generative AI, agentic models required rule-based programming or advanced machine-learning, which made deployment complicated and laborious.
With generative AI, agentic systems can now be built on models that have been trained on massive and unstructured data sets – known as foundational models – rather than pre-defined rules. This gives them the ability to adapt to different scenarios.
What can agentic AI do for CX?
Agentic AI clearly has the potential to revolutionize process management and operations, as well as shared services and finance. In fact, almost every industry and sector will benefit from automated, autonomous assistants that can devise and conduct tasks to reach a specific goal or outcome. As a result, customers will benefit, too.
In CX, the new capabilities could be explored and applied in a number of different ways, whether that involves back-end processes, data collection or, the really obvious one, supporting agents (more on that below).
However, as the ultimate autonomous assistant, it is also possible for agentic AI models to execute other tasks. For example:
Training: If a CX leader needs to train their team on a new system or process, agentic AI could proactively research the topic, create a syllabus, distribute the training materials then relay the agent’s progress and feedback.
Marketing: While generative AI can facilitate hyper-personalization by creating endless versions of the same advert to push to small customer cohorts and individuals, agentic AI could analyze business objectives, customer data, VoC insights and social media trends to continually refine the targeting of various campaigns and test them across user groups and platforms.
Profit protection: UK-based Pets at Home is an early adopter of Microsoft Copilot Studio, which allows organizations to build AI assistants for employees and customers. It has used the technology to create an agent for its profit protection team so it can more efficiently compile cases for skilled human review. The retailer says this could have the potential to drive a seven-figure annual savings.
Machine customers: Hotly tipped to be the next big thing since ChatGPT was unveiled, fully autonomous machine customers are now a step closer to reality. Because Agentic AI can devise and complete tasks, where generative AI could previously draft a letter of complaint or handle a service interaction, agentic AI can now initiate an interaction, send that letter and handle the follow-up.
As exciting as these possibilities are, the application of agentic AI must be managed. If bots become the sole executors of mundane, process-related work and agents are expected to handle only technically and emotionally complex scenarios in service, Jeannie Walters says there is a hidden danger of agent burn out.
She explains: “Empathy is a finite resource. We have to make sure that we are proactively thinking and designing the workplace and the way agents work in a different way. Because right now, those little mundane tasks actually provide a break as they don't require all that emotional energy.”
Who is providing agentic AI solutions?
The first movers on the agentic AI trend include Microsoft. Its Copilot Studio provides an all-in-one platform for building agents, with managed software as a service (SaaS) infrastructure, AI models, a low-code design interface and thousands of prebuilt connectors. The technology is set to debut at Microsoft Ignite in November 2024.
There’s more on Microsoft’s technology in this video
OpenAI, the creator of ChatGPT, has released Swarm, which it describes as “An educational framework exploring ergonomic, lightweight multi-agent orchestration”. Although some may find the idea of a “swarm” of AI agents unnerving, OpenAI says the new tool “provides AI developers with a means of trying out agentic AI capabilities”. Furthermore, these are “lightweight” agents, hence the need for the swarm.
In October 2024, agentic AI developer Materia was acquired by Thomson Reuters, which is executing on its own AI roadmap to provide generative AI tools to the professions it serves. Materia is a US-based startup that specializes in the development of an agentic AI assistant for the tax, audit and accounting profession and its agentic AI assistant automates and augments research and workflows, helping accountants to improve efficiency, effectiveness and the value they add to their clients.
Freshworks launched its solution in October 2024 as an extension of its Freddy AI suite. Intended to deliver stronger customer and employee experiences, Freddy AI Agent can be deployed in minutes and has helped users in customer support and IT autonomously resolve an average of 45 percent and 40 percent of service requests, respectively. Brands such as Bchex, Porsche eBike, Hobbycraft and Live Oak Bank are already using this agentic AI solution and have reported higher CSat scores and reduced response times as a result.
Built on generative LLMs, Freshworks' Freddy AI Agent is platform-agnostic and follows the June 2023 generative AI launch of Freddy Self Service (now Freddy AI Agent), Freddy Copilot (now Freddy AI Copilot), and Freddy Insights (now Freddy AI Insights).
Twilio announced a developer preview of its AI Assistants back in April. It’s the first project launched under the Twilio Alpha program and is described as “a platform to build customer-aware autonomous agents that we call Assistants.” In Twilio’s own words, AI Assistants can leverage tools defined by the user to make API requests to interact with other systems and access knowledge sources to tap domain expertise without a human having to manage their own RAG pipeline. If an Assistant can't answer a question or the organization doesn't want it to handle certain topics, AI Assistants can automatically hand over the conversation to a human agent in the Twilio Flex contact center.
You can find out more in this video
Can agentic AI take my job?
As we learned when generative AI burst onto the scene, the idea that AI can replace the work of humans doesn’t go down too well with the majority of humans, whether they’re agents or customers.
Still, this hasn’t stopped some organizations from starting to replace their human workers with AI. In mid-2024, Klarna’s financial results confirmed the buy now pay later (BNPL) giant had reduced headcount by more than 1,000, with plans to make further reductions as it cut costs ahead of a planned IPO.
Yet the EU AI Act, which came into force shortly after this news broke, mandates that organizations using AI chatbots must be transparent in how they do so, and give customers the chance to refuse a bot interaction in favor of a human interaction.
In addition, article 50 of the Act states that those using generative AI to create content must disclose that the text has been artificially generated or manipulated, unless the content “has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.”
When unveiling its agentic AI solution, Microsoft has clearly stated its vision is to allow one agent to do the work of many, therefore safeguarding at least some jobs, although the IMF has previously said that in advanced economies, about 60 percent of jobs may be impacted by AI.