Every wave of automation in education has produced the same headline: the machines are coming for teachers. It has never quite happened that way. What changes is the work around teaching — and right now, AI agents are changing it faster than most job descriptions.
When an agent can draft feedback, answer routine questions and chase late submissions, the scarce skills move elsewhere. If you are building a career in edtech, that is where to look.
The roles taking shape
- Learning engineer — combines learning science with building. Designs how an AI tutor should behave, then tests whether learners actually learn more.
- AI workflow designer — maps a team's processes, decides what an agent should do and what stays human, and writes the rules in between.
- Evaluation designer — builds the test sets and rubrics that tell a team whether an agent is getting better or worse. Few people do this well, and every serious team needs it.
- Mentor or facilitator — the human role that grows rather than shrinks. With the admin handled, the job becomes coaching, motivation and judgement.
Skills worth building now
1. Specifying work clearly
Agents are only as good as the goal and rules they are given. Writing a precise, testable description of a task — inputs, steps, edge cases, what "done" looks like — is quickly becoming a core professional skill.
2. Evaluating AI output
Anyone can judge one answer. The valuable skill is judging a hundred of them systematically: building a sample, defining what good looks like, and spotting patterns in the failures.
3. Understanding how people learn
Retrieval practice, spacing, feedback timing, motivation. Teams that pair agent-building with real learning science build products that work; teams that skip it build very fast ways to hand learners answers.
4. Enough technical fluency to build a prototype
You do not need to be a machine-learning engineer. You do need to be comfortable connecting a model to a spreadsheet, a document store or an LMS, and reading the results critically.
How to show it in an interview
Credentials help less than evidence. Bring one small thing you built and can explain end to end: the problem, the agent's tools, what went wrong, how you measured it and what you changed. A two-page write-up of a working prototype says more than a list of courses.
The strongest portfolio piece is not the most impressive demo. It is the one where you can show what failed and how you found out.
The busywork is going to the agents. The judgement — what to teach, how to know it worked, when a learner needs a person — is staying with people. Build a career around that.
Frequently Asked Questions
Do I need to learn to code for a career in AI-powered edtech?
It helps, but it is not the entry point for most roles. Clear task specification, evaluation and learning design matter more for workflow and learning-engineering roles. Basic scripting and comfort with APIs make you noticeably more effective.
Which edtech roles are least likely to be automated?
Roles built on judgement and relationships: mentoring, facilitation, curriculum decisions and evaluation design. Agents are taking over the repetitive tasks around those roles rather than the roles themselves.
What should a first portfolio project look like?
Something small and real: an agent that drafts study plans from a syllabus, or one that turns quiz mistakes into revision sets. Document the goal, the tools it uses, how you tested it and what you changed afterwards.
