Enterprise AI Tutors: What L&D Leaders Should Know

Only 15 percent of the students who could use Khan Academy's AI tutor ever bothered to open it. That number came from Khan Academy's own team this month as it rebuilt Khanmigo from scratch, and it should stop every L&D leader who just watched Workday ship an AI tutor to the enterprise. Enterprise AI tutors are now generally available, vendors promise up to three times the engagement of legacy training, and whether yours gets used or ignored comes down to one design decision most teams get wrong.

Here is what an enterprise AI tutor actually is, what the research says about whether it works, why so many go unused, and how to build one that people actually turn to.

What is an enterprise AI tutor?

An enterprise AI tutor is a conversational learning assistant built into a company's training platform that answers questions in plain language, explains concepts tied to the material in front of the learner, and adapts to their role, skills, and progress.

Think of it less as a search box and more as a coach that lives inside the course. A learner stuck on a compliance scenario can ask "why does this rule apply to my region?" and get an answer grounded in the actual module and their own job context, not a generic web result.

The difference from a plain chatbot matters. A chatbot answers whatever you type. An enterprise AI tutor is scoped to the content, the learner, and the outcome the business cares about: can this person now do the thing the training was meant to teach.

Why enterprise AI tutors are suddenly everywhere

They are everywhere because the biggest HR platforms just made them a default feature instead of an add-on. On July 22, 2026, Workday made Workday Learning, powered by Sana, generally available, putting a personal AI tutor in front of every employee on the platform.

The pitch is aimed straight at L&D pain. According to Workday, customers reported up to a 98 percent reduction in content creation time, five times faster compliance reporting, and three times higher learner engagement compared with legacy systems. The tool turns existing PDFs and slide decks into structured courses in minutes and translates them into dozens of languages.

The framing is telling. Workday Chief AI Officer Joel Hellermark put it bluntly: "Checking the box isn't the same as building a skill." Industry analyst Josh Bersin said the approach converts large program libraries into interactive courses "in months rather than years."

Those numbers are vendor-reported, so treat them as marketing until your own pilot confirms them. But the direction is real: the AI tutor is becoming a standard layer of the corporate learning stack, not a science experiment.

Do enterprise AI tutors actually improve learning?

A well-designed AI tutor can come close to a human tutor, but the research is clear that it does not automatically beat one, and a badly designed one does nothing at all.

Start with the benchmark. In 1984, education researcher Benjamin Bloom described the two-sigma problem: students taught one-to-one by a skilled human tutor outperformed classroom peers by two standard deviations, a massive gap. Decades of technology have chased that number. A widely cited 2011 meta-analysis by Kurt VanLehn found that step-by-step computer tutors reached about 0.76 standard deviations of improvement, close to the 0.79 measured for human tutors, but far short of Bloom's two-sigma ideal.

The newer evidence is encouraging but hedged. Researchers reviewing the current wave of AI schools found no clear evidence that AI or other computer-based tutors are superior to human tutors, though they are often cheaper, as science outlet Phys.org reported. A 2020 National Bureau of Economic Research review found human tutoring produced consistent gains across subjects and ages. A 2025 Harvard physics study did find students learned faster with an AI tutor than in class, but those students were already motivated and the tutor was built by their own instructors.

The honest read for L&D: the technology can work, the ceiling is high, and the outcome depends almost entirely on design and adoption rather than on the model itself.

Why most enterprise AI tutors get ignored

Most AI tutors get ignored because they sit next to the content instead of inside it. That is the lesson Khan Academy learned the hard way with the first version of Khanmigo.

Only about 15 percent of eligible students actively used it, per Khan Academy's internal data reported by The Learning Standard. Chief Learning Officer Kristen DiCerbo was direct: "Too many students who had it available did not even try it." Founder Sal Khan admitted the first version "did not change student learning as much as many of us hoped."

The diagnosis is the important part. Khan's conclusion, as covered by EdTech Innovation Hub, was that "the AI could not just sit next to the content. It had to be woven into it." A helpful assistant in a sidebar is easy to close and even easier to forget. If using the tutor is an extra step, most people skip it.

The rebuilt Khanmigo flips the model. It is visible during the work itself, it starts the conversation by asking learners to explain their reasoning, and it is designed to prevent cognitive offloading by pointing to where a learner might slip rather than handing over the answer. In corporate terms: it makes people practice, not copy.

How to make an enterprise AI tutor stick

To make an enterprise AI tutor stick, stop treating it as a help desk bolted onto slide decks and design the interaction into the learning itself. Five moves separate the tutors people use from the ones they close.

1. Weave it into the content, not beside it. The tutor should live inside the lesson a learner is already doing, tied to the exact concept on screen. A tab someone has to remember to open is a tab they will not open.

2. Make it proactive. Do not wait for a question. Prompt the learner to explain their thinking, check for understanding at the right moment, and step in where people usually get stuck.

3. Design for practice, not answers. The goal is a skill, not a completed module. Push learners to attempt, get feedback, and try again, so the tutor builds competence instead of enabling cognitive offloading.

4. Ground it in real role and skills data. Tutoring that knows a learner's job, region, and prior progress can give relevant guidance instead of generic explanations. This is exactly why platform-native tutors tied to HR data are gaining ground.

5. Measure the outcome, not the chat volume. Track whether learners can perform the task afterward, not how many messages they sent. Engagement is a means, capability is the goal.

This is the same principle behind interactive training videos: the questions, feedback, and adaptivity live inside the lesson, not in a sidebar you have to remember exists. When the interaction is the experience rather than an optional companion to it, the 15 percent problem largely disappears.

FAQ

What is the difference between an enterprise AI tutor and a chatbot?

A chatbot answers any question you type, with no memory of what you are learning. An enterprise AI tutor is scoped to specific training content and to the individual learner, using their role, skills, and progress to guide them toward a defined outcome. The tutor's job is not to answer trivia, it is to build a skill.

Can an enterprise AI tutor replace human trainers?

Not on current evidence. Research consistently shows AI tutors can match or approach human tutors under good conditions, but there is no proof they beat skilled humans, and the strongest results often come from blending the two. The realistic play is augmentation: let AI handle scale, repetition, and in-the-moment questions so human experts focus on coaching, judgment, and edge cases.

How do you measure whether an AI tutor is working?

Measure downstream performance, not activity. Look at whether learners can complete the real task after training, whether time-to-competency drops, and whether performance holds up weeks later, not just how many messages the tutor exchanged. High chat volume with no capability gain is a warning sign, not a win.

Enterprise AI tutors are about to be everywhere, but availability is not adoption and adoption is not learning. The teams that win will be the ones that stop bolting a chatbot onto passive content and instead build the interaction into the experience, so every learner is thinking, practicing, and getting feedback rather than watching. That shift, from passive to interactive, is what turns a shiny new tutor into training people actually remember.