will AI replace tutors?

The work that makes tutoring actually work is relational. When a student shuts down because they're embarrassed they still don't understand fractions, you notice. You slow down. You change your tone. You tell them about the time you got it wrong too. AI can't read a 12-year-old's body language across a kitchen table, and it can't earn trust over six sessions the way you can.

The tasks with zero AI penetration tell the story clearly. Identifying that a student needs a specific intervention plan, writing an individualised education plan, preparing a tutoring session around that one student's gaps and goals, these all require judgment that sits in your head, not in a model. You're synthesising what you observed last Tuesday, what the parent told you by email, and what the student's teacher flagged. That's not a task you can hand off.

Providing feedback is the same. Positive reinforcement isn't just saying 'good job'. It's knowing when this particular student needs encouragement and when they need a harder push. Communicating progress to parents takes tact. A progress report isn't just data, it's a conversation that requires you to manage expectations, maintain the relationship, and sometimes deliver difficult news. According to O*NET task data, every single one of these relationship-and-judgment tasks sits at 0% AI penetration. That's not a coincidence.

AI tutoring tools hallucinate. When a student asks a maths question and the tool gives a confident wrong answer, the student doesn't know it's wrong. That's the specific risk. A child learning long division from a tool that makes an arithmetic error in its worked example isn't just confused, they're being taught incorrectly. You catch that. The tool doesn't know it made a mistake.

There's also the accountability gap. If an AI tool gives a student wrong advice about their college prep timeline or misreads a learning disability, nobody is responsible. You are. That accountability structures how carefully you work. AI tools have no skin in the game and no liability. They can't be held to the same standard, and most of them aren't designed to flag their own uncertainty in ways that a ten-year-old can interpret.

Privacy is a genuine issue too. Platforms like Khanmigo collect conversation data from minors. Parents and schools are increasingly asking who owns that data and how it's used. You're bound by professional and legal expectations around student confidentiality. AI platforms are bound by terms of service, which is a very different thing. That gap creates real institutional hesitation about replacing human tutors with AI tools for vulnerable or young students.

Four tutoring tasks are now genuinely AI-accessible, and you should know exactly what that means. Scoring assessments is the clearest one. Tools like Gradescope can mark short-answer and multiple-choice work in seconds, flag common errors, and return results faster than any manual process. If you're spending 40 minutes a week marking practice tests, that time is recoverable.

Reviewing class material and providing practice instruction is where AI tutoring tools are most active. Khan Academy's Khanmigo can walk a student through a maths problem step by step, ask Socratic questions, and adjust based on whether the student got it right. Synthesis Tutor is built specifically for problem-solving practice with children. These tools are genuinely useful for repetitive drilling work, the kind of thing a student needs to do 20 times before it sticks but doesn't need a human for every repetition. That's a real capability, not hype.

Developing teaching materials is the third area. Tools like Curipod and MagicSchool AI can generate quizzes, worksheet drafts, and study guides from a topic or a reading level in under two minutes. Recommending learning resources is faster too. A quick prompt to an AI tool can surface five textbook options with grade-level summaries in the time it used to take to search a catalogue. These are time savers for prep work. They don't touch the session itself.

The biggest shift is in prep time. What used to take 30 minutes, building a worksheet, finding practice problems at the right difficulty level, drafting a progress note for a parent, now takes five to ten minutes. You're arriving at sessions more prepared, not less, because the grunt work of assembly is faster.

The session itself hasn't changed. You still sit with the student, you still watch their face, you still rephrase the explanation three different ways until one lands. That part is identical to how it was five years ago. What's changed is what happens before and after. Less time on materials. Less time on admin. More time thinking about the specific student in front of you.

What hasn't changed at all: the scheduling, the parent calls, the moment a student finally gets something and you have to decide whether to push further or let that win sit. None of that is different. The job has more support around the edges, but the centre of it, the human interaction that makes tutoring worth anything, is exactly what it always was.

The BLS projects 0.6% growth for tutors through 2034. That sounds flat, and it is. But flat growth on a base of 215,500 employed tutors, with 37,100 annual openings, means the field is replacing people and staying stable, not shrinking. The openings number is what matters most if you're thinking about job security. Demand isn't collapsing.

The honest tension is that AI tools are filling some of the lower-stakes tutoring work. A parent who needs their kid to drill vocabulary for an hour doesn't need to hire you for that anymore. Khanmigo and similar tools are handling that end of the market. Where you're protected is the higher-need student: the one with a learning difference, the one whose confidence is shot, the one whose parents are paying for a real relationship and real accountability. That market isn't going away.

According to BLS occupational data, tutoring sits in a category that's growing alongside broader demand for education support services, even if the headline growth number is modest. The Anthropic Economic Index classifies tutoring in the 'amplified' quadrant, meaning AI is changing the work without eliminating the worker. The roles that survive in this quadrant are the ones where you use AI to do more in less time, not the ones where you try to compete with it on the tasks it's already taken.

The AI exposure score for tutoring is 54%, and it's likely to hold or creep upward over the next five years as AI tutoring tools get better at adaptive learning. The current tools are decent at drilling and decent at explaining. They're not good at motivation, relationship management, or handling a student in genuine distress. For those capabilities to be automated, you'd need AI that can read emotional states reliably in real time and respond in ways students actually trust. That's at least ten years away, and probably more.

The realistic near-term pressure is at the lower end of the tutoring market: basic homework help, test prep drilling, and simple concept review. Those tasks will increasingly shift to AI tools. Where the role is safe is in the work that requires you to know a specific student over time, adapt to their specific gaps, and carry the relationship with their parents. If AI tutoring tools improve dramatically in the next decade, the question isn't whether you'll be replaced. It's whether you'll need to position yourself clearly as the high-need, high-relationship option rather than the general-purpose one.

The tasks to double down on are the ones with zero AI penetration: building intervention plans, running progress conversations with parents, and developing individualised session plans. If you can get trained in learning difference support, IEP development, or specific methodologies like structured literacy or Singapore maths, you're moving into territory that AI tools can't touch and that commands higher rates.

On the admin side, start using the documentation tools covered above. If you're still writing parent progress reports from scratch or building every worksheet by hand, you're spending time on tasks that have already been automated. Recovering that time lets you take more students or go deeper with the ones you have. That's a real competitive advantage over tutors who haven't made the shift yet.

The career move worth thinking about is specialisation. Generalist tutoring, helping any student with any subject, is the part of the market most exposed to AI tools. Specialist tutoring, working with students who have dyslexia, preparing students for specific high-stakes exams like the SAT or AP courses, or supporting students with IEPs in school settings, is the part that's growing and that AI is worst at. The tutors who'll feel the most pressure in the next decade are the ones doing low-stakes, repeatable subject review with no particular specialism. The ones who'll be fine are the ones who've built a reputation, a method, and a client base around specific hard problems.