Kian Katanforush had one of the best mentors in the AI world: renowned researcher Andrew Ng, who also served as Katanforush's advisor at Stanford University's graduate school. The two launched Stanford's deep learning program, and Ng is now chairman of Katanforush's startup, Workera. Ng was a key mentor in Katanforush's career, and now, as CEO of Workera, he's trying to figure out what makes a good mentor and automate it through his new AI agent, Sage.
“I trust Andrew because I understand his background and his expertise, but how many people like Andrew are there in the world? There aren't many,” Katanforouche said in an interview. “So automating this aspect of mentorship was really important.”
On Tuesday, Workera introduced Sage, a conversational AI agent designed to assess employees' skill levels, goals and needs. Workera says that after taking a few short tests, Sage accurately assesses how proficient an employee is in a particular skill. Sage can then recommend appropriate online courses through Coursera, Workday or other learning platform partners. Through a chat with Sage, Workera will test employees' current writing, machine learning and math skills and provide a path to improvement.
Sage (Workera) Screenshots
To be clear, Workera's definition of a “mentor” is quite specific. Katanforoosh acknowledges that Sage can't do everything Ng did for them: encouragement, career guidance, and networking are hard to automate. But Sage can provide a somewhat objective assessment of an employee's skill set and recommend appropriate courses to help them achieve their goals. It's not a perfect mentor, but it's better than what some people have access to.
Katanforoush is the son of Iranian immigrants whose parents were forced to flee their country during the destabilizing revolution of the 1970s, abandoning their education in the process. His father abandoned a science degree and began selling clothes in France to make a living. While Workera currently targets mostly employees of Fortune 500 companies, Katanforoush believes that by making its assessment skills more accessible, it could one day help people in the same situation as his parents.
People crave mentorship today more than ever before, and in the age of remote work, junior employees have less in-person time with more experienced colleagues, meaning fewer opportunities to pick up valuable wisdom around the water cooler. Workera's CEO thinks its new AI agent could be up to the task.
Sage is scheduled to roll out to early access customers, including military contractor Booz Allen, starting in November 2024. Other Workera customers, including the U.S. Air Force and Accenture, will have general access to Sage in March 2025.
Workera has raised over $44 million to provide employers with a way to administer AI-generated tests to their workforce and evaluate their employees' skill sets. Employees don't typically welcome the opportunity to be compared to their peers, but Workera is also looking to give companies a way to invest in their employees.
Employer Perspective with Sage (Workera)
Sage does the same thing, but with a more conversational experience that ties into the Workera platform more cleanly. The flexibility of the multimodal models from OpenAI that Workera uses also gives it a flexible interface that can scale to significantly more tasks in different mediums. Katanforoosh says that 95% of Sage's interactions are performed by GPT-4o, and 5% of the time the AI agent uses OpenAI's new o1 model to plan more complex tests that may require inference.
Can AI agents replace good human mentors?
With Sage, Workera aims to become a mentor for employees, not just a manager of skills assessments. A human mentor offers the kind of emotional support, encouragement, and connection that an AI chatbot can never provide. But Katanforoosh believes there are some aspects of mentorship where Sage excels.
Sage provides benchmarks to show progress across different skills (Workera)
“A good mentor needs to give a good assessment. If a mentor can't assess accurately, they can't help you. This is something that can be automated,” Katanforoosh says. “In fact, I'm convinced that today's measurement systems are better than most. When it comes to measuring someone's machine learning skills, I trust the Workera system more than I trust myself.”
Another aspect is that humans are biased and are heavily influenced by superficial characteristics, so they cannot always accurately evaluate a person's talent. Incidentally, AI systems are not perfect either and contain many of the same biases as the humans who created them. After all, AI models are based primarily on human-generated data.
But bias in AI models may have a more promising solution than human bias. Katanforush teaches a course on AI debiasing at Stanford University. He firmly believes there are ways to use algorithms to mitigate bias hidden in data from AI models. These algorithms can weight gender, race, and other considerations differently, making the output of AI models fairer.
“AI is already less biased than humans, but I am confident it will become even less biased in the coming years,” Workera's CEO said.
Catanforush said automating these tasks frees up managers to manage the human aspects of mentorship that AI can't automate: Human managers need to encourage and coach employees, one of the things AI mentors are still limited in how well they can do.
One thing Sage doesn't do yet is teach long-form content — for that Workera is turning to partners in the online learning space — but Sage can identify skills that are likely to be mastered quickly and generate simple scenarios and questions to test comprehension.
One could say that Workera is stretching the word “mentor” here and using a bit of its own definition. That said, Sage could be a useful agent that managers can add to their tool belt for evaluating and investing in their employees.