Guest post by Adonis Celestine, senior director of Applause.
2025 will be the year of hyper-specialised Gen AI chatbots
Up until now, generative AI chatbots have catered to a wide audience, who use them for a range of tasks like drafting emails, writing code, and conducting research. In 2025, we will start to see more hyperspecialised chatbots enter the market that are tailored to industries like healthcare, telecommunications, law, travel, etc.
Unlike the generic Gen AI we are familiar with today, hyperspecialised Gen AI chatbots need to be trained on domain-specific data created by experts in each field, who also have the knowledge needed to gauge response accuracy and quality. This is important because, particularly in sectors like healthcare, where incorrect diagnoses or advice could have grave consequences, chatbots need to know what they are talking about.
The good news for brands is that it’s actually less intensive to train a Gen AI chatbot to perform very well in one specific area than to perform generally well across limitless topic areas and use cases.
Businesses find Gen AI doesn’t ‘stack up’ as legacy infrastructure hinders adoption
Over the last couple of years, many companies launched Gen AI products that have not delivered value at best and completely failed at worst. In the rush to adopt the new technology, companies simply didn’t spend enough time properly thinking their product through. They also didn’t consider how difficult it is to get Gen AI’s entirely new ecosystem to integrate with their legacy IT infrastructure.
Data silos, compatibility issues, and change management challenges are causing problems. As with many projects that are often driven from the top down by company management, so much emphasis is put on speed to market that very serious concerns are overlooked. Little attention is given to ‘minor’ details like compatibility with existing systems or even thorough analyses of the use cases that could have the most impact on the business. Companies rolling out Gen AI products need to put more effort into the planning stage if they want to see a real return on investment.
Automation gets a new lease with agentic AI in 2025
Despite its promise to streamline work, cut costs, and free up time, automation has not really delivered the productivity boost people imagined. While Gen AI has already driven automation forward through its code creation ability, it could not execute actions autonomously on the user’s behalf. This will all change with agentic AI, which integrates with existing systems to perform actions.
The combined power of generative and agentic AI means we can now automate entire processes end-to-end. Imagine you are having a conversation with a virtual health assistant: besides providing diagnostic information, soon it may be able to book a doctor’s appointment for you, organise a taxi to the hospital and make a note to follow up on your recovery. The challenge in 2025 will be to integrate this technology with legacy IT stacks and upskill technical staff to navigate its complex architectures.
In 2025, AI testing other AIs will become more mainstream
You read it right: 2025, Gen AI will test Gen AI. The reason for this is that Gen AI is very difficult to test using traditional QA methods. Gen AI is nondeterministic, meaning that its responses are always unique and therefore impossible to predict. This complicates testing because some responses may pass tests while others fail.
Large language models (LLMs) function on billions of parameters that control things like the relationship between words, contextual meaning and grammar, all of which need to be tested — which is logistically impossible for resource-constrained QA teams. Gen AI, on the other hand, has far greater computational power. Through techniques like benchmarking, where it compares Gen AI responses to example responses curated by humans, Gen AI can check all parameters in seconds. Techniques like these will become commonplace next year.
