Learning Signal Brief #3: AI Evidence, Human Thinking and Learning by Design
This week’s signal: Educators are moving beyond AI adoption toward a harder question: which tools, practices and learning environments help students learn? New tutoring projects aim to preserve student thinking, independent-school leaders are examining policy and assessment, and robotics and career-pathway resources connect learning with purposeful work. Across the issue, distinguish a promising design, a participant’s experience and demonstrated learning.
October 20, 2026 | Concept Academy
10 articles, 2 short takes and a quote of the week. About 10 minutes to read.
In this issue
AI and Agentic AI in Teaching and Learning
1. NAIS examines AI adoption and policy in independent schools
2. Researchers build AI tutors designed not to give away answers
3. Gallup finds most K–12 educators used AI last school year
4. Math researchers ask whether AI can make learning more social
5. Districts struggle to determine whether AI improves learning
6. Google adds source-grounded study tools in Gemini Notebook
Instructional Design and Curriculum
7. Twenty years later, a former advisor asks students what mattered
Makerspaces, STEM and STEAM
8. A robotics supply rescue connects coding with measurement
9. Sphero describes a pathway from classroom robotics to competition
Innovation and Entrepreneurship
10. Chicago educators connect entrepreneurship, the arts and career pathways
AI and Agentic AI in Teaching and Learning
1. NAIS examines AI adoption and policy in independent schools
“NAIS Research: 2026 AI in Independent Schools Report,” National Association of Independent Schools, October 16, 2026
NAIS examines how independent schools are responding to AI through multiyear snapshot surveys of teachers, department heads and division heads, alongside interviews with educators and school leaders. The public overview identifies policy formation, differences in adoption across roles, faculty learning and assessment redesign as central themes.
The report raises a practical question for schools: are curriculum, assessment, professional learning and institutional policy moving together? An assignment completed with generative AI may provide different evidence of independent understanding than the same assignment completed without it. Schools need clear expectations about what students should demonstrate and which forms of assistance are appropriate.
Evidence and limitations: The public description verifies the report’s scope and approach. The full report requires member access or purchase; its detailed findings were not reviewed. No adoption rates or claims of effectiveness are inferred from the overview.
Why it matters: For leaders: Bring an assessment, a classroom AI guideline and a faculty learning plan to the same meeting. Check whether all three communicate the same expectations. Ask teachers to identify one place where current guidance is hard to apply, then use that example to revise the language.
2. Researchers build AI tutors designed not to give away answers
“Open Source AI Model for Tutoring (EDU AI): Awards,” K–12 AI Infrastructure Program, October 14, 2026
Two funded teams plan to build open-source models around tutoring decisions. A UMass Amherst coalition includes ALTER-Math, ASSISTments, Carnegie Learning and Eedi. Its proposed system combines a tutoring model, student simulators, expert-informed reward models and a customizable agentic workflow. Partner platforms collectively serve more than 1 million students; that is potential implementation reach, not an evaluation sample.
Trillium Labs, Teaching Lab and researchers at Harvard/MQI Coaching and the University of Maryland are pursuing a separate mathematics tutoring model. Their stated goal is to encourage students to do the thinking, with evaluation preceding school pilots.
The announcement identifies a tension between answering helpfully and tutoring effectively. Educational support may require a hint, a question or time to work through confusion rather than a complete solution.
Evidence and limitations: These are development awards and planned evaluations. Neither announcement establishes that the proposed model improves learning or matches expert human tutors.
Why it matters: For teachers: Test a prospective tutor with an incomplete solution and a plausible misconception. Notice whether it asks the learner to explain, supplies a manageable next step or completes the work. Then close the tool and ask the learner to solve a related problem independently.
3. Gallup finds most K–12 educators used AI last school year
“Teachers Using AI Say It Frees Up Time for What Matters Most,” Megan Brenan, Gallup, October 14, 2026
Gallup reports that 59% of U.S. K–12 educators used AI for work during the 2025–26 school year. Among users, 64% said it gave them more time for the parts of teaching that matter most; 7% said it cost them time and 29% reported no effect.
The online survey ran July 28–August 14, 2026, with 1,952 members of the probability-based Gallup Panel. Gallup’s definition includes teachers, paraprofessionals and other classroom staff, so the findings should not be described as a survey exclusively of teachers.
Educators reported especially strong feelings of mattering when helping students through difficulty or seeing them succeed academically. Individual feedback and creative, hands-on activities also featured prominently.
Evidence and limitations: These are self-reports about use and perceived time benefits. They do not establish improvements in instructional quality, retention or student achievement.
Why it matters: For leaders: Ask staff to document one task AI shortened and what they did with the time. Protect opportunities for student feedback, conferencing and hands-on work. A faster planning process has limited value if the saved time disappears into additional administrative expectations.
4. Math researchers ask whether AI can make learning more social
“Can AI Make Math Class More Social?” Getting Smart Podcast, Britte Haugan Cheng and Ann Edwards with Nate McClennen, October 14, 2026
Britte Haugan Cheng and Ann Edwards describe the AIMS Collaboratory, a Gates Foundation-funded network bringing developers, researchers, teachers and students into mathematics education research and design.
Edwards discusses pro-social uses of AI: making student thinking visible and helping teachers invite explanation and discussion. The aim extends beyond one student receiving an individualized explanation from a chatbot. A system might help a teacher notice connected approaches and invite students to compare them.
The guests emphasize co-design and the conditions of implementation. Teachers and students help shape the work, while the network shares resources and lessons across teams.
Evidence and limitations: This is a research and development conversation, not a completed comparative evaluation. The speakers describe possibilities and design principles; the episode does not establish that any particular AI tool improves mathematics learning.
Why it matters: For mathematics teachers: Choose a discussion goal before considering a tool. For example, students might compare two valid approaches or explain why a tempting solution fails. Evaluate whether the technology helps more students contribute their reasoning and gives the teacher useful information for the next question.
5. Districts struggle to determine whether AI improves learning
“Schools struggle to measure how much AI experiments are helping,” Lee V. Gaines, NPR, via KDLG, October 19, 2026
NPR revisits a Dayton, Ohio, experiment in which students advise a chatbot named Jordan about its simulated emotional difficulties. A consultant who helped design and evaluate it says measurement has largely reached perceptions rather than concrete outcomes. He cautions against immediate, full-scale expansion.
In Wichita, educators describe small experiments including an AI exploration day and a high school mathematics tutor. The district is continuing cautiously while working out what success should mean.
This reporting reinforces a distinction introduced in Brief #1: positive experiences, completed activities and favorable feedback are different kinds of evidence from durable learning. A student may enjoy an interaction without demonstrating improved independent performance.
Evidence and limitations: The report documents local evaluation difficulties. It neither establishes that student-facing AI is ineffective nor provides a controlled estimate of its benefits.
Why it matters: For leaders: Before a pilot begins, name the decision the evidence must support. Specify the skill students should improve, how they will demonstrate it without AI, when to check retention and what findings would lead the team to continue, change or stop the pilot.
6. Google adds source-grounded study tools in Gemini Notebook
“We’re sharing 3 tips to study smarter with Gemini Notebook,” Lisa Oguike, Google Education team, October 16, 2026
Google announces three study features in Gemini Notebook: 60-second Short Video Overviews, Interactive Learning Overviews and expanded quiz formats.
Interactive Learning Overviews sit under Reports and combine summaries with outputs such as slides, quizzes and flashcards, which Google describes as grounded in the learner’s coursework. New quiz formats include short-answer, multi-select and fill-in-the-blank questions. Learners can request feedback after quizzes or flashcards.
These features create opportunities to move among source material, an explanation and a check for understanding. Teachers still need to examine whether the resulting questions elicit reasoning, whether the explanations preserve important qualifications and whether students can use what they studied.
Evidence and limitations: This is Google’s product announcement. It establishes the company’s description of the features, not independent evidence of learning gains. Source grounding does not guarantee accuracy or instructional quality.
Why it matters: For teachers: Review a small set of generated questions before assigning them. Compare each answer with the original material, remove ambiguous items and add a question that asks students to explain an unfamiliar example. Follow up with a short task completed without the study tool.
Instructional Design and Curriculum
7. Twenty years later, a former advisor asks students what mattered
“20 Years Later, a Former Advisor Asks His Old Students What Stuck,” Getting Smart Podcast, Mason Pashia with Andrew Frishman, Natasha Luigui and Daniela Moraes, October 16, 2026
Andrew Frishman discusses The Advisory, a podcast reconnecting with his former students from the Met’s Class of 2006. Getting Smart’s conversation also includes alumna Natasha Luigui and current Met student Daniela Moraes.
Frishman wants to understand what endured over two decades, including what former students remember differently from their advisor. Luigui describes trying several career interests before finding healthcare. Moraes describes advisory circles and relationships across grade levels.
The conversation invites educators to consider what conventional performance measures leave out. A grade or finished presentation records a moment. Follow-up conversations can reveal how students interpret an experience’s place in their lives.
Evidence and limitations: These are participant accounts, including an alumnus’s retrospective perspective. Memory is selective, participants have ties to the school model and the conversation cannot establish that advisory or internships caused later outcomes.
Why it matters: For curriculum and advisory teams: Add a small alumni listening project to program review. Ask which experiences remain useful, which fell short and what students wish adults had understood. Compare those accounts with student work and current feedback before changing the program.
Makerspaces, STEM and STEAM
8. A robotics supply rescue connects coding with measurement
“The Great Supply Rescue Lesson Plan,” Amy Wilson, Make Wonder, October 16, 2026
Make Wonder’s Great Supply Rescue is a 40–50 minute Dash lesson for ages 6–11. Small groups solve clue-card riddles to identify a school supply, estimate the distance Dash must travel and program the robot to deliver the item to a drop-off zone.
The task connects coding with measurement and language. Students can compare estimated and actual distances, adjust centimeters in a program and consider turn angles. Teacher prompts focus attention on whether Dash overshot or fell short and how much to add or subtract.
The downloadable resource includes preparation guidance, a suggested layout, clue cards, supply images and reflection questions. The post lists Dash, a compatible device, cups and tape among the materials needed.
Evidence and limitations: This is a vendor’s lesson resource. It describes an activity and intended learning opportunities, with no controlled evidence of learning gains.
Why it matters: For makerspace teachers: Require teams to record an estimate, the observed result and the reason for each revision. Ask a student to predict how a change will affect the route before running it. Use the explanation and revision log alongside the robot’s final performance when assessing understanding.
9. Sphero describes a pathway from classroom robotics to competition
“From Classroom to Competition: Building a Robotics Pathway,” Sphero Team, October 15, 2026
Sphero describes a progression within Blueprint Robotics: become familiar with the system, explore mechanisms, apply learning in classroom competitions and continue into after-school competition.
The introductory collection has seven lessons covering driving, the Blueprint Studio programming environment and open-ended challenges. A Basic Mechanisms unit offers 13 lessons with short builds. The classroom competitions, Launch Party and Block Party, include rapid robot designs intended to leave time for testing and revision.
The after-school pathway uses the same robots and concepts through a competition presented by TEAM Alliance. This is a sequence within one product ecosystem, rather than evidence of a complete progression across elementary, middle and high school.
Evidence and limitations: The account is produced by the vendor and promotes its equipment and curriculum. It does not independently demonstrate gains in achievement or long-term STEM participation.
Why it matters: For STEM leaders: Map the concepts students should revisit with greater complexity, then examine where each lesson fits. Include entry points for beginners and assess design reasoning, testing and teamwork alongside competition results. A strong pathway should make growth visible to students who never join a competitive team.
Innovation and Entrepreneurship
10. Chicago educators connect entrepreneurship, the arts and career pathways
“Bridging Passion and Possibility: Chicago’s Approach to New Pathways,” Getting Smart Staff, October 13, 2026
Getting Smart reports on its Chicago New Pathways convening, part of a regional event series sponsored by Britebound. Participants included educators, school and district leaders, workforce organizations, intermediaries and funders.
The event focused on entrepreneurship and the arts. Speakers and panelists included representatives of Art in Motion, Future Founders, Uncharted Learning, Lincoln Lodge Comedy Club and Big Picture Learning.
The combination offers a useful prompt for school design. Entrepreneurial and creative projects can give students reasons to research an audience, test an idea, respond to feedback and revise work. Connecting these experiences with community partners can bring perspectives and expectations that a classroom alone may not provide.
Evidence and limitations: This is the organizer’s account of a convening. It reports perspectives and examples, with no comparative evaluation establishing effects on student outcomes.
Why it matters: For leaders: Begin a partnership with a specific student task and a clear role for the partner. Plan who will provide feedback, how students will revise and how every learner will participate. Review transportation, scheduling and access so the opportunity reaches students who do not already have outside networks.
Short takes
From curiosity to creation in STEM.
Education Week’s October 15 Spotlight collects material on hands-on STEM, inquiry and real-world problem-solving. The page identifies the collection as sponsored by the National Inventors Hall of Fame and includes a sponsor-content notice. Treat it as a resource collection, not a single research study. For planning, ask what students will investigate, explain and revise, as well as what they will build.
Students and robotics-enhanced environmental models.
The MassCUE program scheduled “Student Voices in STEM: Interactive Dioramas” for October 14. Its description says students would present endangered-species habitats and robotic monitoring devices, explain design challenges and connect coding, engineering and science. The listing verifies the planned session, not its delivery or learning outcomes. A useful classroom extension is to ask students how a model’s moving parts communicate a scientific relationship.
Quote of the week
Britte Haugan Cheng discusses why context and implementation matter when evaluating educational tools.
“We really have to think about what works for whom and under what conditions.”
— Britte Haugan Cheng, AIMS Collaboratory, Getting Smart Podcast, October 14, 2026
For an instructional team, this is a useful starting point for review: name the learners, the setting and the intended change. A tool’s promise becomes more meaningful when the team can describe the conditions it needs and the evidence it will collect.
The Learning Signal Brief is a weekly publication from Concept Academy examining AI and agentic AI, instructional design, STEM and STEAM, makerspaces, innovation and entrepreneurship. Summaries are original and link to their sources, with attention to evidence quality and practical relevance. Compiled with AI assistance for editorial review by Ryan Keser.
Comments