For decades, in schools and workplaces, we’ve relied on static assessments: fixed questions, fixed scoring, and little room for nuance. Most of these tests are binary. You get the answer right, or you don’t. That’s a poor measure of capability!

In real work, the process matters as much as the result. I want to see how someone thinks about a problem, how they break it down, where they go off-track, and what they do when they hit uncertainty. That information tells you whether someone can grow into a role, lead a project, or tackle harder challenges.

But with binary scoring, you lose all of that. Even when a learner is “almost there,” the test still says they failed. That’s the difference between “not there yet” and “completely lost,” and traditional assessments rarely capture it.

Why conversation changes the signal

Conversational assessments change the type of signal we collect. By talking through a problem, the person reveals their approach, reasoning, and blind spots. You can see where they’re strong, where they need support, and how close they are to mastery, even if they arrive at the wrong final answer.

This isn’t new. We’ve been using conversation for evaluation for centuries (oral exams, technical interviews, mentorship sessions, etc.). But those methods don’t scale. No manager or mentor can run deep, one-on-one assessments with every employee on a regular basis, especially in large, global, multilingual teams.

Introducing Sage’s ability to Talk

I’ve been fascinated with speech models since graduate school at Stanford. Many of you might remember the “trigger word detector” from the Deep Learning Specialization or Stanford’s CS230, which I taught with Andrew Ng. That’s why I’m especially excited about this update to Sage.

With guidance from the OpenAI team, our engineers set out to solve three hard problems in parallel:

  • Creating a natural, human-like voice experience
  • Building a model that can perform tool calling accurately and score conversations reliably
  • Achieving low latency, even under heavy load, so it feels fluid at scale

The result is Talk: Sage’s voice mode for running skill-based challenges in conversation. It adapts in real time, adjusting difficulty and probing further based on demonstrated skill. It can award partial credit (which is extremely difficult for today’s models), making follow-up learning plans more precise and preventing near-ready talent from being overlooked.

Talk breaks down assessments into skill-based challenges rather than isolated questions, so learners can complete them in smaller pieces while still maintaining psychometric validity. It works today in English, Spanish, French, and German, with more languages on the way. It’s powered by our skills ontology and Deep Knowledge Tracing model, which means it’s not just a “prompt-based chat experience” — it connects to psychometrically sound scoring and role-specific learning recommendations.

It’s also fast: average response time is under 1,200 ms from speech to follow-up generation, so the interaction feels natural, almost like speaking to a human mentor. We’re still working on reducing latency, and it’ll get much better soon.

Scaling effective, unbiased assessments

Introducing conversational assessment isn’t just about improving the effectiveness of employee learning programs. It’s about access, fairness, and meritocracy.

We designed Sage’s voice capability specifically to feel like a trusted AI mentor. The world needs more mentors, particularly in places that have historically had fewer opportunities for learning and access to technology. Talk makes it possible for companies to scale mentorship, particularly in different languages where they may not have as many internal leaders and mentors.

Sage’s new voice capability also helps to enhance fairness in skills assessments and hiring. We know that bias is a bug that all humans struggle with; even those who have gone through dedicated anti-bias training can have “bias blind spots.” By putting assessments (and eventually, staffing decisions) in the hands of neutral, unbiased AI interviewers, we make it possible for organizations to establish true meritocracies.

The organizations that win in the future will be those who can identify the best people and help them reach their full potential. With conversational assessments, Sage now makes it easier to get there.