We’ve all interacted with a chatbot in one form or another –– those cheerful little popups that promise instant help and too often deliver canned responses. It’s no surprise that many of us have defaulted to mashing “live agent!” into the chat window just to get something done. Yet conversational AI remains one of the most hyped areas of generative AI development—and one with enormous potential, if we can get it right. In fact, according to Forrester Research, in 2023, 71% of companies were either experimenting with or implementing conversational AI and chatbots on their websites.
So how do we reconcile these things –– and what’s all the excitement about?
Hype doesn’t always translate to impact. I’ve seen firsthand how flashy AI features can miss the mark when they don’t fit into real-world workflows. The real promise of conversational AI isn’t novelty—it’s utility.
The Challenges of Traditional Chatbots
Until recently, most chatbots were developed over a limited domain. Each task that a company wanted to perform––or knowledge that you wanted the chatbot to have––had to be custom developed. Imagine, for example, a customer service chatbot trying to arrange for an item to be returned. It must have the capacity to handle issues regarding the reason for return, validation, customer history, return policies, shipping, logistics, and finance. All of that to arrange the simple return of a sweater.
As we get to more complex tasks that require multi-step problem solving and interactive feedback from the customer, it gets even harder. These are just some of the reasons that while chatbots remain “cool,” we still tend to ask to speak with a live human when we actually need to get something done.
That’s the disconnect. If AI doesn’t solve a problem—or if it’s not easy to use in the context of your actual workflow—it’s not much more than a shiny object. Our job as technologists is to cut through the hype and build tools people actually want to use. That also means making AI work within the systems and experiences teams already rely on, not expecting users to adapt to AI.
Key Developments in Conversational AI
A set of recent developments could be changing a lot of this, opening conversational AI up to be more powerful, reliable, and able to take on tasks of increasing complexity. The overall performance of LLMs, the increasing ability and understanding for LLMs to step through a problem, and the emerging field of Agentic AI and standard protocols around it could all have major implications.
As providers build larger, more powerful and more reliable LLMs, their ability to handle increasingly complex tasks––as well as formulate a stepped plan––continues to improve. Overall, we can expect a steady rising tide in all places where conversational AI is being used thanks to these changes. In addition, with the overall LLM models and ability to follow guardrails improving, we can also expect fewer hallucinations and a corresponding increase in reliability.
Conversational AI is getting smarter. A recent study published in Scientific Reports evaluated how well conversational AI—using seven major LLMs including GPT4, Gemini, Llama 3.1, and Mixtral—can detect nuanced meaning in text: sentiment, political leaning, emotional intensity, and even sarcasm. Researchers compared AI performance to 33 human coders across 100 curated messages and found these models performed on par with humans in most areas. GPT4, in particular, showed special strength in identifying political bias consistently. It also accurately gauged emotional intensity and sentiment, though it tended to underrate the strength of emotions. Sarcasm detection, however, remained challenging for both humans and AI, with no system clearly outperforming people.
These findings point to a future where AI can support—not replace—human analysis, especially in domains where nuance, sentiment, and subjectivity matter. But the path forward must be paved with careful evaluation, not blind trust.
The implications are significant. For fields like social science, public health, journalism, and content moderation, AI can dramatically speed up analysis of large volumes of text—flagging emotionally charged or politically slanted content in real time. Still, researchers caution around bias, transparency, and the consistency of AI judgments under varying prompts. More work is needed to ensure stability and fairness before deployment in sensitive domains.
While these advancements are exciting, they come with responsibility. Large language models are excellent at sounding confident, even when they’re wrong. That’s why a focus on accuracy is so essential –– especially in high-stakes legal and business workflows. Getting the right answer is not a nice-to-have; it’s essential. And too often, that’s overlooked in the rush to roll out new AI features.
The Rise of AI Agents and Agentic Protocols
What is most exciting is the rise of AI Agents and Agentic protocols. AI Agents allow a conversational AI to break down a complex task into multiple sub-tasks, formulate a plan, and critically access a variety of tools and purpose-built AIs to accomplish them. Equally exciting, the rise and increasing adoption of standard protocols like Model Context Protocol (MCP) allow for an emerging ecosystem of AI Agents and related tools. This means that a single company or developer doesn’t have to build every single piece of their conversational AI, but can leverage capabilities built by others, much as we leverage APIs and tooling integrations today.
For example, a hotel company can build an AI agent to handle finding and booking rooms . This agent can then be used by a travel company as part of an overall vacation planner that includes flights, hotels, car rentals, and activities. Each sub-agent can be built once by various companies and used by anyone. All of this can lead to a single chat experience where a customer can reliably get personalized vacation advice, price a vacation acceptable to their budget, and have the chatbot make all arrangements for them, down to ordering special meals on their flight.
The Expanding Possibilities of Conversational AI
As the conversational AI ecosystem continues to grow, its potential to handle larger, more complex, and more diverse tasks is expected to increase significantly. With the introduction of new tools and subagents, and as more users become proficient in leveraging them, the emergence of conversational AI systems capable of autonomously managing nuanced and sophisticated tasks through natural dialogue will become more common.
In many ways, we’re still in the dial-up era of conversational AI. The potential is there, but the fundamentals still matter—especially in contract lifecycle management, where precision and trust are everything. We’re not chasing magic; we’re building tools that deliver clear, consistent value. Because if AI isn’t solving real problems, it’s not the future. It’s just a demo.
