AI and teacher workload have become closely connected in discussions about educational technology. A new systematic review by Danny Delgado-Togra and Grethy Quezada-Lozano (2026) suggests that automation redistributes teaching tasks and responsibilities.
Some operations become quicker, while teachers take on work involving verification, interpretation, privacy, and oversight. Whether this improves teaching or saves time depends on how the technology is introduced and supported.
I find this a useful starting point because it brings professional judgement into the discussion of efficiency. A tool can produce a lesson plan or feedback quickly. Teachers still need to decide whether that output is accurate, appropriate, and useful for particular learners.
The work that remains after an automated response
Consider a teacher using an automated system to draft feedback on student writing. The draft is only one part of the task. Someone must check whether the comments address the learning goal, represent the student’s work fairly, and offer a realistic next step. Errors require correction; unsuitable advice requires revision.
This example is my illustration of the review’s broader concern: automation changes how work is distributed. Evaluating a tool therefore requires attention to the whole teaching process, including the decisions that follow its output.
That question belongs within the wider discussion of AI and teaching practice. It also explains why teachers’ hesitation deserves closer examination. The authors suggest that resistance can reflect inadequate support, uncertain accuracy, or recommendations that conflict with professional experience. Labelling every concern as unwillingness to innovate can obscure genuine implementation problems.
Summary of the study: AI and teacher workload
Delgado-Togra and Quezada-Lozano reviewed 58 empirical studies indexed in Scopus and Web of Science and published between 2020 and 2024. They followed PRISMA reporting guidance for systematic reviews and assessed methodological quality using the Mixed Methods Appraisal Tool.
Their definition of educational automation included AI, learning analytics, adaptive platforms, robotics, and automated feedback. This breadth matters: the paper cannot be read as an evaluation of today’s generative AI tools alone.
The synthesis identified connected patterns around adoption, the redistribution of teaching work, and educational outcomes. Teacher competence, institutional support, and the conditions for professional decision-making were recurring concerns.
The authors interpret the changing role as involving design, mediation, curation, verification, and ethical supervision. They also discuss potential benefits for feedback, engagement, learning, and inclusion, while explaining that these depend on teaching design, accessibility, infrastructure, and governance.
The evidence supports qualified conclusions. Many studies used observational designs, self-reports, or short interventions, and some measured student outcomes without directly examining changes in teachers’ roles. The review was restricted to English and Spanish publications with retrievable full texts, and initial screening was not fully duplicated.
A sensitivity analysis retained the broad themes but reduced confidence in some outcome claims. Consequently, the review provides a framework for examining change; it does not establish that automation generally reduces workload or improves educational quality.
Practical insights: Evaluating the whole teaching workflow
Measure the work surrounding the output. The researchers recommend assessing workload before and after implementation. I would apply this to a bounded task, such as preparing feedback for one assignment.
Record the time spent producing, checking, correcting, and delivering it. Also examine the quality of the feedback. A faster draft is useful only if the full process remains manageable and educationally sound.
Specify which decisions remain with teachers. The authors recommend clear boundaries around partially automated tasks and professional decisions. Learning objectives, assessment judgements, accommodations, and sensitive feedback require direct professional control.
A school adopting a system should establish who checks its outputs, how errors are handled, and how teachers can override recommendations. “Human oversight” needs a practical meaning in the daily workflow.
Make professional learning about judgement. The review draws on TPACK, which connects technological, pedagogical, and subject knowledge. Training should give teachers opportunities to test tools against actual teaching demands and discuss accuracy, bias, and suitability. This connects with my earlier review of Intelligent-TPACK and ethical AI integration. Knowing which buttons to press leaves substantial pedagogical work unresolved.
Include accessibility and support in the pilot. The researchers treat inclusion as something implementation must address and evaluate. Access to a platform alone tells us little about whether learners can use it meaningfully. A local pilot should examine accessibility, connectivity, available human support, and whether particular students encounter additional barriers.
I would also make teachers’ ability to reject an output visible during professional learning. Participants could explain what they changed, what they kept, and why. This is a suggested adaptation, not an activity tested by this review, and it connects naturally with evaluative judgement in the AI age.
The review leaves me with a concrete question for technology adoption: after a task is automated, who does the remaining work, and do they have the time, expertise, and authority to do it well?
Reference
Delgado-Togra, D., & Quezada-Lozano, G. (2026). Transformation of the teaching role through educational automation and intelligent technologies: A systematic review. Frontiers in Education, 11, Article 1876239. https://doi.org/10.3389/feduc.2026.1876239
