Journal for School Superintendents · Evidence into practice

Agentic education.
The science of learning with AI.

Explore what tutors, adaptive systems and AI agents can help learners do—and what must still be demonstrated without assistance. Read the licensed research, inspect its population and methods, then open an age-sensitive implementation companion.

634new source papers and companions

3,634 / 7,000overall verified CC source collection

5Agentic & multi-agent systems47Conversational & pedagogical agents134Intelligent tutoring & adaptive learning34Embodied learning agents414Supporting AI learning science

Explicitly agentic systems are labeled separately from chatbots, adaptive tutors and supporting AI learning science. Original authors and DOIs are preserved. Automated relevance, license and design screening is not an independent quality appraisal. Companion guides are AI-assisted editorial material and are not counted as additional source papers.

Different learners. Different evidence requirements.

Stage tags come from source titles and abstracts; exact participant ages require full-text verification. Papers may mention multiple stages, including in reviews or teacher-perception studies; these counts are not counts of child-participant trials. A higher-education finding does not establish effectiveness for children, and an age-band count does not establish a proven intervention.

Stage-specific research questions and collection coverage
Stage mentioned in sourceDesign focusQuestion to testPapers mentioning stage
Early childhood · preschoolAdult-mediated language, play and turn-takingDoes the child demonstrate the skill with a person after the tool is removed?24
Primary schoolGuided hints, concrete examples and fading scaffoldsCan learners explain and transfer the idea independently?17
Secondary schoolSelf-regulation, reasoning and confidence calibrationDoes AI support thinking or replace the learner’s first attempt?72
Children · stage unspecifiedDevelopmental context must be verifiedAre participant ages and the educational setting established?59
K–12 · mixed / unspecifiedGrade-specific appraisal before implementationWhich grades were actually studied, and do results differ by age?19
Higher educationDisciplinary reasoning, feedback and assessmentAre gains measured in students rather than in model exam scores?297
Adult & professional learningAuthentic practice and expert-supervised simulationDoes performance persist on a new task without AI?37
Age / stage not establishedPopulation verification before implementationCan participant ages or education stage be established in the full text?196

What the research currently supports

K–12: inspect the comparator

A systematic review of 28 studies involving 4,597 K–12 learners reported generally positive results for intelligent tutoring, with smaller advantages against non-intelligent tutoring comparators. The included studies were quasi-experimental and varied in duration. This supports carefully designed pilots, rather than an assumption that any AI tutor will improve learning.

Read the review and search its evidence →

Preschool: individual differences matter

A study of 28 children aged 4–5 followed four robot-supported word-learning sessions. Shy children showed different interaction behavior and lower initial post-test scores, but the reported gap closed on a later test. This is a small study of individual differences within robot interaction, not a demonstration that robots outperform teachers. It argues for examining delayed outcomes and learner variation rather than equating expressiveness with learning.

Read the preschool word-learning study →

Young learners: separate engagement from attainment

A nine-week elementary afterschool robot-theater program involved 38 children aged 6–10; 21 participated in research activities. The study reports perceptions and enthusiasm, alongside teacher concerns about distraction and peer relationships. It does not establish a randomized improvement in academic achievement or justify extrapolation to preschoolers.

Read the child–robot study →

University: promising effects can have weak certainty

A review of 66 randomized trials with 4,911 undergraduate health-professions learners found that most studies had high risk of bias and rated certainty low to very low. Positive knowledge, satisfaction or confidence results must be read alongside those limits. These are adult disciplinary settings, not evidence of effectiveness for school-age children.

Read the RCT evidence synthesis →

These brief interpretations are AI-assisted editorial synthesis of the linked source abstracts, not an independent systematic review. Read each full paper’s methods, outcomes and limitations before making an implementation decision.

Performance is not learning

Separate immediate assisted task performance, student satisfaction and model benchmark scores from delayed retention and unaided transfer.

Agency requires a boundary

Specify goals, tool permissions, memory, supervision and stopping rules. A chatbot label alone says little about autonomy or instructional quality.

Evidence needs a comparator

Ask who learned, compared with what, over how long and under which conditions. Review sample size, attrition, outcome validity and risks of bias in the full paper.

Research, agents and actionable companions

414 matching papers. Design labels are screening aids; verify methods before drawing causal conclusions. Clear filters

Supporting AI learning science

Large language model-generated versus teacher-written objective structured clinical examination stations for medical students: a blinded comparative pilot study

Piotr Szychowiak; Jonathan Wong-So; Hélène Messet; Mélanie Faure; Isaure Breteau; Simon Jamard; François Barbier; Maxime Desgrouas

Journal of Educational Evaluation for Health Professions · 2026-05-26 · CC BY 4.0

Higher education
Model benchmark / tool evaluation

Assessment & reasoning

Read sourceActionable companionPaper agent

Supporting AI learning science

Education Research: Quality of Narrative Feedback Generated by a Large Language Model Compared With Expert Faculty for Case-Based Learning in Neurology Education

Hannah Fruitman; Sasha Severin; Atikul Miah; Christina Gao; Haelynn Gim; Carolyn Qian; Sang-O Park; Kelly Hou; Edward L Kong; Benjamin Cook; Jasmin Le; Brandon Stretton; John Maddison; Liam G McCoy; Luke Collins; Andrew Vanlint; Rudy Goh; Matthew Arnold; Aye Thant; Rani Priyanka Vasireddy; Doris Kung; Ashley M Paul; Haatem Reda; Tamara B Kaplan; Adam Karp; Galina Gheihman

Neurology: Education · 2026-05-27 · CC BY 4.0

Age / stage not established
Design / feasibility study

Feedback & scaffolding · Assessment & reasoning · Accessibility & inclusion

Read sourceActionable companionPaper agent

Supporting AI learning science

Development of a Clinical Clerkship Mentor Using Generative AI and Evaluation of Its Effectiveness in a Medical Student Trial Compared to Student Mentors: 2-Part Comparative Study

Hayato Ebihara; Hajime Kasai; Ikuo Shimizu; Kiyoshi Shikino; Hiroshi Tajima; Yasuhiko Kimura; Shoichi Ito

JMIR Medical Education · 2025-09-04 · CC BY 4.0

Higher education
Evidence synthesis

Feedback & scaffolding · Assessment & reasoning · Accessibility & inclusion

Read sourceActionable companionPaper agent

Supporting AI learning science

Multiagent Large Language Model Framework for Psychotherapy Fidelity Assessment in Motivational Interviewing and Cognitive Behavioral Therapy Training: Cross-Sectional, Simulation-Based Evaluation Study

Mohammad Amin Kamaleddin; Mina Mirjalili; Reza Barzegar; Nghia Trung Le; Zachary Cote; Olga Winkler; Lisa Burback; Yanbo Zhang; Osmar R Zaiane; Candice Monson; Divya Sharma; Sri Krishnan; Andrew J Greenshaw; Richard Zeifman; Peter Selby; Venkat Bhat

JMIR Medical Education · 2026-08-18 · CC BY 4.0

Age / stage not established
Evidence synthesis

Feedback & scaffolding · Motivation & engagement · Assessment & reasoning

Read sourceActionable companionPaper agent

Generative AI education leadership collection · Full leadership collection