Back in May, we held a webinar unveiling our top 10 emerging technologies for 2025. We had a very strong turnout for the interactive event and received a lot of insightful questions from attendees. In fact, we received so many questions that I couldn’t answer them all during the live event. So I took some time to draft responses to those we didn’t get to during the webinar and have posted them here for everyone to read.

And by the way, if you’d still like to watch the webinar, it’s not too late — the recorded version is available here on demand. Now, on to the questions.

Would you consider a humanoid robot a sophisticated “physical” application of generative AI models?

That’s an interesting thought. Humanoid robots are definitely starting to use language models, but I would not call them an application of language models. Interaction in natural language has been a long-standing requirement and a difficult one to meet for earlier designs. Humanoid robots are using more of the emerging reasoning capabilities in advanced foundational models to help them physically respond to external stimuli, but this advancement is new. Check out Gemini Robotics.

As more processes are being automated by AI, do you think that there will be too many humans for the number of jobs available in five years?

Forrester tracks these trends closely, and my colleagues Michael O’Grady and J. P. Gownder will have an updated forecast on this very soon to reflect the latest changes. But in general, we think that there will be some net job loss due to AI, but we also think that AI will create many new jobs, roles, and even entirely new industries. Certainly, there are many implications for companies and workforces based on these trends, but perhaps the biggest one is that every worker and every company needs to improve their AI quotient (AIQ), which measures how ready they are for AI.

What do you see as the most common pitfalls of synthetic data for digital health?

Overall, the most common pitfall is expecting synthetic data to do something it was not designed to do, and that would apply to digital health as well as any use case. The production of the right synthetic data sets for each of the use cases I covered in the webinar is as much art as science, often requiring many iterations to ensure that the data set can accomplish the expected goals. Synthetic data to help de-bias the data for a predictive model will be quite different than knowledge distillation data used to train smaller versions of generative models, for example.

How does the projected timeline for quantum security to reach its benefit horizon compare to how fast quantum is progressing?

We expect quantum computers to have roughly a 33% chance of breaking today’s PKI encryption by 2035, based on the latest consensus of quantum computing researchers. By implementing all the elements of quantum security over the next two to five years in a phased rollout, enterprises will have a high degree of protection, if no engineering breakthrough accelerates this quantum timeline. As a side benefit, the cryptographic agility you get from quantum security will also help your entire security posture because you will be equipped to replace crypto that is vulnerable to good old-fashioned hacking.

What are the biggest pain points and benefits for agentic AI use by small- and medium-sized enterprises?

In business overall, agentic AI is most used for employee support use cases due to limited trust — looking up data from multiple sources and synthesizing it into a useful answer is most common. TuringBots are also using more agentic AI to automate the software development lifecycle. In consumer businesses, agentic systems are being offered by startups such as Genspark that can plan travel, book restaurants, and even build websites, but these are early and mostly “toys.” Marketers are experimenting with agents that can offer personalized shopping recommendations or build rich content messages. The common thread for all of these use cases is that they are low-risk but also of moderate to low benefit. It is going to take some time to develop the trust and security infrastructure needed for critical business-process steps and decisions or high-impact customer interactions.

When it comes to agentic AI, trust is key. How can we sufficiently test agentic AI enough to trust it to be put into production?

The simplest answer is that there is no known way to align today’s foundational models to ensure that they always do what we want them to. This is called the alignment problem, and it is a grand challenge in all AI. The key question for most enterprises is: How good is good enough? If you must be right 100% of the time, an agentic system won’t get you there. But you probably aren’t 100% correct and reliable today with any software system you use, and certainly, human employees are prone to errors, as well. Since you already know how to deal with errors created by software and human workers, the challenge is selecting use cases, defining error thresholds, and testing the agentic AI systems for those so you can live with (and trust) the agents in specific contexts.

How do you see agentic AI and the agentic web emerging in the future?

The “agentic web” is a term being floated about to describe a world where humans interacting with businesses (and each other) on the web are replaced by AI agents interacting with each other on behalf of those humans and businesses to carry out tasks such as information search and completing transactions. Today, this is more concept than reality. As envisioned, it will be more open and fragmented, likely disrupting the entire search and e-commerce industries. But to realize this vision, we need to develop and implement the right standards and do more foundational work. Some of this has started with protocols like the Agent2Agent (A2A) protocol and the Model Concept Protocol (MCP), but many others are needed (most notably, a model authentication protocol). How will a company know that an agent is truly representing a real person, customer, or partner company? How will that agent properly assert its authority to act on behalf of its principal? It is too early to know, but it’s a great callout, and these trends definitely bear watching.

Next Steps

Thanks again for all the great questions. If you’re interested, watch the on-demand version of the webinar here and download our guide to the top 10 emerging technologies of 2025.

And one last callout: If you’re planning on coming to our upcoming CX Summit North America later this month, I’ll be presenting a session entitled “Emerging Technology Dissolves The Barriers Between You And Your Customer,” which will cover some of these topics from a CX perspective. There will also be sessions covering AIQ, AI agents, and many other emerging tech topics, so be sure to check out the agenda here.