Should everyone have an AI teammate?

With talk of agentic AI and digital twins gripping the industry, it can sometimes feel like a company not already automating workflows and implementing “AI teammates” is living in the past.

But it’s still early days, and companies are figuring out the best use of these latest innovations. For some, it’s simple: What if executives were available to answer questions and talk with employees without needing to actually participate in a conversation? Or even better, what if everyone in the company had a digital twin that was just like them, only trapped inside the cloud?

To get a sense of how far this AI teammate tech has actually progressed, I set up a demo to see how they worked. I was also able to talk to an AI persona I never expected: one of myself.

Personal AI, not Kamino. Personal AI, a company that builds out AI teammates, envisions a time when every person has their own personal AI. Specifically, it is aiming to create digital twins that can act as a knowledge base, answering specialized questions for other users.

Jonathan Bikoff, chief business officer at Personal AI, told IT Brew that the company has embraced an “ensemble of approaches” composed of distinct “small language models,” retrieval-augmented generation, and “some LLM access.” A user can hook the model up to specific digital folders or labels on platforms like Gmail or Google Drive, Bikoff said.

The utility of a small language model depends on how much data the user feeds it. If the AI teammate lacks data, it will respond to questions from a user with a response along the lines of “I don’t have the answer,” or it will retreat to using a third-party LLM. (Bikoff said that he was unable to name the LLM that Personal AI uses “because customers bring their own. Sometimes it’s self-hosted or cloud-hosted or on-prem.”)

Persona-lity. The persona that a user or organization builds out has a combination of long-term, short-term, and relationship memory. Bikoff said during a demo of the model that “each user has the ability to create and access and manage their personas or shared personas that they’ve been added to.”

“There’s a lot of configurability in terms of instructions and guardrails and limitations and behaviors and guidance and all this stuff, as well as directives,” Bikoff said. “I can effectively tell your persona to behave differently or act differently, depending on its audience.”

No, I’m the real Caroline Nihill. Bikoff and I met via videoconferencing, where he walked me through how this product works, pointing out the ways that someone can train a persona on available data—in my persona’s case, it was trained on articles I’ve written.

After our discussion, I was granted the ability to speak with my digital twin. Because it was trained on my publicly available articles, if I asked it a question outside the realm of tech journalism, it would simply say, “I cannot answer that or don’t have anything relevant in my memory. I’ve let Caroline know.”

But when I asked the bot if it had spoken directly to Caroline, it insisted that it was Caroline Nihill, a tech reporter at IT Brew: “Let me be clear, I am Caroline Nihill…I apologize for my earlier inconsistent response, which may have caused confusion.”

Later, it changed its mind, identifying itself as my AI persona, “trained to assist with technology reporting and analysis for IT Brew,” adding: “Let’s focus our discussion on enterprise technology, cybersecurity, and IT strategy topics that I cover in my reporting.”

From a privacy perspective, one might find having a digital twin to be an unnecessary risk to take during a time where cybersecurity is at the forefront of most organizations' concerns. Given the endless warnings that bad actors are looking for any footing to impersonate an executive, convince others to hand over access, and deceive vulnerable systems, I'm doubtful about the helpfulness of this tech.

From up high. Digital twins are already here and transforming a variety of industries, according to a study by Strategic Market Research, which found that 75% of businesses are already investing in digital twins in some capacity. And the digital twin industry is expected to keep growing—McKinsey & Company reported last year that the global market for digital twin tech will grow 60% over the next five years, reaching $73.5 billion by 2027.

Some IT pros might find the idea of a helpful assistant that can retrieve data in their absence useful, as well as something that could give them peace of mind if someone has a question for them after office hours. But will this reporter want a digital twin anytime soon? All signs point to no.