Kian Katanforoosh has one of the world's best AI mentors: renowned researcher Andrew Ng, who also happened to be his graduate school advisor at Stanford. The two went on to help create Stanford's deep learning program, and now Ng serves as chairman of Katanforoosh's startup, Workera. Ng had been an amazing mentor in Katanforoosh's career, but what the Workera CEO wants to do is break down and automate what it means to be a good mentor through a new AI agent, Sage.

"I trust Andrew because I understand his background and expertise, but how many Andrews are there in the world? Not that many," Katanforoosh said in an interview. "So automating this mentorship aspect was critical."

On Tuesday, Workera launched Sage, an AI agent you can talk to that can assess a job candidate's skill level, goals, and needs. Workera says that after some quick tests, Sage will accurately place someone on some continuum of proficiency for any given skill. Then, Sage can steer them towards relevant online courses available through partners like Coursera or Workday. Workera is said to meet employees where they are, chatting with Sage, testing their skills in writing, machine learning, or math, and giving them a path to improve.

For those familiar with Workera, Katanforoosh assures readers that Sage is not to be mistaken as a mentor, meaning someone like Ng who would do everything and more for him. Mentoring is an incredibly tough thing to automate: encouragement, career advice, networking. But Sage can offer the employee a partial objective assessment of their skills and recommend the appropriate courses to acquire such skills. That's hardly enough for a good mentor, but it is better than many have at their disposal.

As the son of Iranian immigrants, Katanforoosh's parents had to leave their home country during the unstable 1970s, leaving unfinished studies. His father abandoned a degree in science to sell clothes in France in order to survive. Today Workera services mainly Fortune 500 employees, but Katanforoosh sees making skills in assessment available one day help out people like his parents.

More than ever, people are craving mentorship today. While remote work and virtual collaboration has spread to become widespread in the workplace by limiting young employees' face time with experienced colleagues, it limits those opportunities for catching some nuggets of wisdom by the water cooler. That is why the CEO of Workera believes its new AI agent is up for the challenge.

Sage will be rolled out to early access customers by November 2024, including defense technology provider Booz Allen. Other Workera customers, including the U.S. Air Force and Accenture, will get general access to Sage in March 2025.

Workera has successfully raised more than $44 million to administer AI-generated tests for the employees of the enterprises, and in doing this is providing employers with a way of measuring the skills of employees. Employees do not usually jump at the chance to be ranked with their fellows, but Workera is trying to give business models an avenue through which to invest in their employees as well.

Sage does the same thing but offers a much more conversational experience that ties the Workera platform together much more neatly. It also has a flexible interface built from the multimodal models used by OpenAI that is capable of scaling to drastically more tasks in various mediums. Indeed, 95% of Sage's actions are running on GPT-4o, says Katanforoosh; the AI agent runs OpenAI's newest o1 model just 5% of the time in order to blueprint out more elaborate tests that involve some form of reasoning.

Can an AI agent replace a great human mentor?
Workera asserts, via Sage, to be a mentor to the workers rather than just merely the administrant of skill assessments. An autonomous human mentor provides emotional support, encouragement, and contacts that a highly ambitious AI chatbot probably never could. Still, in certain aspects of this mentorship, according to Katanforoosh, Sage can still be improved.

"A good mentor needs to assess properly, because unless the mentor can assess accurately, it cannot help you… That's one that can be automated," Katanforoosh said. "In fact, I'm pretty confident measurement systems we have today are better than most people; I would trust the Workera system much more than I trust myself at measuring someone's machine learning skills."

Another more similarly relevant factor is that humans do tend to be biased, and susceptible to superficial characteristics, so are often wrongly judged on someone's talent for a particular use. AI systems can also be wrong sometimes, for the record, and have many of the same flaws as humans who built them. Most AI models are based on human-generated data.

But the biases of AI models may have a more promising solution than that of humans. Katanforoosh teaches a course at Stanford that covers bias mitigation methods in AI. He firmly believes there are ways to reduce bias underlying an AI model's data with algorithms. These can weight gender, race, or other considerations in a different way, making the outputs of AI models more equitable.

"I really feel very confident that AI is already much less biased, but will be even less biased than humans in the coming years," said Workera's CEO.

However, says Katanforoosh, such automation would allow managers to free up time to focus on such human ingredients like encouraging and guiding their people, among other things out of the reach for AI mentors.

One area where Sage doesn't yet stretch is teaching long-form content. For that, Workera relies on partners within the online learning landscape. However, Sage should point out skills you could learn quickly, then generate a short scenario and question to gauge how much you know.

Workera is really stretching the word "mentor" here and kind of using its own definition. However, Sage may become one useful agent that managers can add to their tool belt to help assess and invest in the workforce.