AI digital twins for everyone
(Gift link gets you past the paywall)
AI digital twins are a fascinating idea and clearly will be challenging for any company to implement legally and ethically.
Eric Yuan, the CEO of Zoom, said he was exploring this in a Decoder podcast interview with Nilay Patel at The Verge a while back.
This opens up so many questions for me as a designer with a degree in Industrial Organizational Psychology.
What contexts does it make the most sense for where and when these AI clones should be integrated at a company?
Is this a benefit exclusive to the C-suite class?
How much does this cost in AI tokens, which are sometimes used to obscure the fiscal costs of AI use?
Is anyone at the company empowered to create their own digital twin to take on their key responsibilities at work?
What business contexts does it make sense for enabling AI to perform such key levels of decision-making?
Gaps in inference by the large language models used behind the scenes of modern chatbots and friendly voice assistants may not always be the best solution, especially knowing how biased and sycophantic they can be to the user.
Clear policies, procedures, and company-wide training customized for each business line would be best for companies who want to embrace this kind of headcount efficiency superpower.
Companies are strongly encouraged to define humans-in-the-loop to prevent people from just running off with an AI output right into key decisions for any organization.
I was asked to provide use cases for AI in the Webroot + Carbonite consumer product user experience, the product managers proposed a couple of options:
A customer support chatbot that generates videos to help people with learning how to set up and use Webroot from product support content.
A chatbot that reduces support calls by automating the top reason for calls: account-related issues with refunds and renewals.
The first one seemed like a good idea in theory, using to generate video content that informs less tech savvy users how to use the product
However, in order to generate videos, someone needs to be good at prompt engineering and have up-to-date written content to generate video content to teach users about the product and how to use it. The challenge would be having enough product knowledge to write accurate, up-to-date support content to structure with AI tools what effectively becomes a lesson plan.
Apple is said to be using this in their sales coach content: https://www.macrumors.com/2026/05/12/apple-sales-coach-will-use-ai-generated-presenters/
The tricky part is this requires actually people to create a plan for an effective content strategy with an ability to maintain and update this, which is now a larger scope request to deliver a learning management system. As the product is updated, the content needs to adapt and change with new features. It’s possible, but sounds like more work for more people to do, not necessarily more efficient in the short term.
Product management decided to pilot a program with the development team to track down product content currently held in a 3rd party content management system and extract as plain content writing instructions for the helper AI agent to follow and content to learn from.
The second option was even more tricky. After validating with our AI engineering team, we were cautioned not to use AI for account related issues. Our engineering team agreed with our UX team that using AI to act on behalf of consumers to drive automation related to accounts is risky, especially with cancellations, refunds, and collecting personal information like credit cards. Seems like a costly and risky move to trust a language model with potentially flawed inference with customer credit card data and automating payments when our business team is already concerned about chargebacks, a costly process with credit card companies when a retailer is seeing a number of customers dispute charges to their credit card.
I provided a rough proof-of-concept AI chatbot user interface based on good ideas in others I saw elsewhere, but the engineering team still had a lot of work tracking down the existing product content and assigning someone to revise it to train the model planning to be used to help this AI helper. This will still require a human-in-the-loop to ensure customers aren’t annoyed or surprised to find recurring charges on their credit cards, or we risk higher costly chargebacks, not less.
I developed other ideas for product and cybersecurity related use cases that could drive more value to the product itself.
Nothing I’m free to share publicly, of course.
Knowing the risks we faced and challenges my team overcame, along with finding opportunities and costs to use AI tools in our UX practice, I can’t help but think about the challenges teams face with making digital twins in other use cases.
For example, who on a Human Resources team owns AI implementation?
It’s obviously a bad idea for an AI tool to inform people their benefits are getting more expensive and their real income is dropping.
Do you trust AI solutions to communicate and interact with people asking personal information about their healthcare needs, compensation, or other challenging employment relationship emrequiremnts?
Or to break the news to long-time trusted employees that they are part of a downsizing decision that the CFO enforced across all teams. This always feels terrible and can even effect people in parts of a business that clearly serve a critical role.
I have been let go from jobs before, once due to a poor training program at a label company and I had to work with an older supervisor who clearly didn’t like my skills with newer versions of software and had a very set way of doing things, once at a startup due to the 2008 market crash, and once due to poor strategic decision making by an online public furniture retailer when Wall Street wasn’t impressed by their lack of profitability and the stock price tanked right as the COVID-19 pandemic hit. How how that information is delivered REALLY makes a difference in how people feel about the severance.
You wouldn't want the perception that your company is an inhuman, uncaring place managed by robots that just fake empathy as they console someone learning they're being let go from a place they worked tirelessly to improve upon.
This can really weigh down an individual, a small team, a large international team, or an entire global corporation trying to do their best under difficult and challenging circumstances.
Using AI so broadly at an organization can create risks to the perception of a company culture.
One has to ask how that impacts opportunities for recruiting talent.
Enabling AI to handle key interactions with people at any level is going to continue to challenge legal, ethical, and cultural norms.
As we see AI adoption in the workplace expand, it will be increasingly critical for people know which tools to use in which contexts it’s appropriate to apply it.
