In 2025, maybe we will begin to prefer AIs over humans when we need help or are trying to learn.
Dr. Joseph Gonzalez, US Berkeley
TLDR
*What are AI agents?
*How can AI agents be used as TAs?
*How can AI agents become teachers?
*AI limitations are overrated
*Adapting to the Future-Present
*Opportunities
*Challenges
*Time to Adapt
Introduction
Last week, there was much focus on “AI Agents.” While such agents have existed for some time, two factors triggered the focus on a more massive deployment in 2025.
(1) Sam Altman’s blog post that AI agents will be “join(ing) the workforce” and
(2) Jensen Huang’s section of his CES Keynote on AI agents and their role in the workforce. That full section is here.
Dr. Sabba Quidwai clipped the section on what AI agents mean for education.
In this post, I’ll explore in more detail what an AI agent is, how agents can be deployed across what I’m loosely calling the entire instructional scheme, and how educators may want to deploy agents to make their lives easier and strengthen instruction. Although robotics is making aggressive strides, the advances I’m discussing are still limited to the digital realm.
What are Agents?
Most people understand AI as generative AI as being text to output (text, image, video). But agentic AI is text to action. At its most basic level, a “Generative AI agent can be defined as an application that attempts to achieve a goal by observing the world and acting upon it using the tools that it has at its disposal.”
Agentic AI systems enhance the capabilities of language models by combining them with advanced techniques for metacognition, external tool manipulation, strategic planning, and multi-agent cooperation. This integration elevates them beyond basic text generation into sophisticated reasoning frameworks that address diverse and complex problems.
Though they are grounded in generative AI, particularly LLMs, they are thought to be the next phase of AI development, and we can already see them in action. . For instance, these models can tap into databases to retrieve specific details like customer purchase histories, enabling personalized product recommendations. They can also interface with APIs to perform tasks like sending emails or executing financial transactions when prompted by users. This functionality requires access to external tools and the ability to plan and execute complex tasks independently. When a generative AI model combines reasoning capabilities, logical processing, and connections to external systems, it becomes an agent - evolving beyond the boundaries of a conventional AI model to engage more directly with the world.
The model used by an agent can be one or multiple LM’s of any size (small / large) that are capable of following instruction based reasoning and logic frameworks, like ReAct, Chain-of-Thought, or Tree-of-Thoughts.
As noted by Huyen, “(i)ntelligent agents are considered by many to be the ultimate goal of AI.” The idea isn’t new. Bill Gates first conceived of agents in his 1995 book, The Road Ahead, which describes an agent as “a filter that has taken on a personality and seems to show initiative. An agent’s job is to assist you.” Gates was writing about agents in the early days of the internet, so he primarily thought of them as research assistants, but we now that agents can write code, answer customer service calls, and help manage human resources inquiries. But, of course, the ability to do research is not insignificant (Accenture 2025).
According to the Accenture Report, agents may become the “primary users of most enterprises' internal digital systems."
To accomplish the goal of assisting you, agents reason, plan, use tools and proof their work.
Jensen Huang, the CEO of NVIDIA, defines agents as AI systems that can act autonomously. They make decisions and act independently based on their environment, goals, and training. Unlike traditional AI operating strictly within pre-programmed instructions or responses, agentic AI can adapt, learn, and respond to situations dynamically.
To break it down, agents are growing increasingly capable in the following areas. Limitations within these will be discussed below, but this is what is coming in the future.
Training
Learning. Agents can learn from examples of work outcomes, receive feedback, and undergo evaluations, similar to how human employees are onboarded and trained. Nvidia has developed an entire system for onboarding AI agent employees — Nemo — that can easily be applied to other areas.
Autonomy. Agents can perceive their environment, reason about it, plan actions, and execute them autonomously. Automated AI agents can set their own tasks and plan out activities. Perception allows them to continually learn and to act on behalf of the user. “Even in the absence of explicit instruction sets from a human, an agent can reason about what it should do next to achieve its ultimate goal.”
Tool Usage. Agents can use tools, reason, and even create tools, which they can test, save, and use for later tasks. Bringing tools to the project allows agents to do the best possible work.
Evaluation. AIs can evaluate their outputs.
Agents that can evaluate their output are more likely to give you what you need.
Collaboration. Agents can collaborate, assemble, and work together to solve complex problems. Collaboration + combining tools increases the complexity of the tasks that can be executed. You can imagine it as a whole team of employees working together plus maximizing their use of the best available tools.
Lu, Pan, Baolin Peng, Hao Cheng, Michel Galley, Kai-Wei Chang, Ying Nian Wu, Song-Chun Zhu, and Jianfeng Gao. "Chameleon: Plug-and-play compositional reasoning with large language models." Advances in Neural Information Processing Systems 36 (2024).
These articles go into much more detail about how agents operate. I ordered them in terms of how useful I found them to be.
Agents: autonomy, self-learning and swarms for the future of AI. (very simple)
Most recently, Chip Huyen’s Agents
Google researchers’ Agents
Anthropic’s Building Effective Agents.
"Balancing autonomy and alignment: a multi-dimensional taxonomy for autonomous LLM-powered multi-agent architectures." arXiv preprint arXiv:2310.03659 (2023).
How is Industry Using Agents?
The most common uses of agents are currently for coding and research (and I use them frequently for this). A couple of days ago, the Wall Street Journal shared some other uses that Connor Grennan summarized
1. Johnson & Johnson — Speeding Up Drug Discovery
Agents turn a manual lab process into an automated system that finds the perfect conditions for creating drug molecules, using digital replicas to test scenarios instantly.
2. Moody's Creating AI Research Teams
Thirty five specialized agents work together like a research team, each bringing different perspectives to analyze company filings and industry trends. My favorite part - they even argue with each other.
3. eBay Agents to Help Buy and Sell
Their agents write software code, create marketing materials, and make the buying/selling experience smoother by learning from user preferences.
4. Deutsche Telekom Building the Employee Help Desk
An agent helps 10,000 employees weekly with everything from company policies to vacation requests, eliminating repetitive tasks.
5. Cosentino Reimagining Customer Service
Their agents handle customer inquiries so effectively they've taken over the workload of several full-time employees, letting their human team focus on complex problems.
Each company is treating agents like team members - giving them specific roles, clear guidelines, and proper oversight.
In his CES Keynote, Jensen Huang outline for examples of domain-task specific expert agents.
Research Assistant Agent: Capable of reading complex documents such as lectures, journals, financial reports, etc., and generating interactive podcasts for easier learning;
Software Security AI Agent: Helps developers continuously scan for software vulnerabilities and prompts appropriate actions;
Virtual Laboratory AI Agent: Accelerates compound design and screening, quickly finding potential drug candidates;
Video Analysis AI Agent: Based on NVIDIA Metropolis blueprints, analyzes data from billions of cameras, generating interactive searches, summaries, and reports. For example, monitoring traffic flow, facility processes, and providing improvement suggestions.
How are Research Scientists Using AI Agents As Assistants?
One example of how research scientists are using AI agents comes from Johns Hopkins University and AMD. They developed "Agent Laboratory," an open-source framework designed to streamline scientific research by combining human creativity with AI-powered workflows. Unlike AI systems that generate research ideas, Agent Laboratory focuses on making existing research processes more efficient. The system works in several stages, beginning with a PhD agent that searches academic papers using the arXiv API. Then, PhD and postdoc agents create a research plan based on the literature review, followed by an ML-Engineer agent that handles technical work using mle-solver for machine learning tasks. Finally, PhD and professor agents write up findings using paper-solver.
In testing, OpenAI's o1-preview model performed best overall, particularly for clarity and validity, while o1-mini excelled in experimental quality. However, AI reviewers consistently rated papers 2.3 points higher than human reviewers. The system can operate in fully automated or co-pilot mode (working alongside humans). While co-pilot mode generally produced better results, it sometimes compromised experimental quality and usefulness. Cost-wise, Agent Laboratory can produce papers for as little as $2.33 using GPT-4o, which offers the best balance of performance and cost. The researchers noted some limitations, including AI's tendency to overrate its work, automated research constraints, and the risk of generating incorrect information.
What are the Current Limitations of AI Agents?
The above description of agents is based on how AI agents will likey evolve, not all of the capabilities they currently possess. Common current limitations of AI agents include.
Needing human validation at critical decision points
An inability to effectively handle novel or unexpected situations without human intervention
A need to follow pre-programmed sequences when using tools
Human oversight or pre-approval for tool use
Human intervention when agents make mistakes or encounter edge cases
A need to operate within strictly defined parameters and decision trees
What Does this Mean for Education?
In Education, how can agents, when given goals, use planning, reasoning, and tools to accomplish objectives?
Let’s examine each stage of what I broadly define as the instructional cycle. I have outlined how AI agents could enhance each phase and sub-step of teaching: Preparation, Delivery, and Follow-Up.
In each section, you’ll see explanations of what AI could do and how the AI’s “agentic capability” (its ability to perform tasks, research, deliver independently, and follow up might operate in practice).
The following section looks at how agent AIs can function as TAs.
The following section looks at how, once they overcome the limitations described above, they will be able to teach or at least should be able to handle a substantial portion of the teaching load. Their abilities in this regard will likely follow a “jagged edge,” perhaps being more factually accurate but less capable of motivating students.
AIs as Teaching Assistants
PHASE 1: PREPARATION
1. Identify Learning Objectives
Review Curriculum Standards
AI Capability: An AI agent could be connected to a specific database of state or national standards (e.g., Common Core, NGSS, local school district guidelines) a particular school system uses. The AI scans your unit/topic and automatically suggests relevant standards.
How It Works:
The teacher enters a topic or learning goal (e.g., “Teach the quadratic formula”).
The AI searches a standards repository and returns the exact standards that match.
AI might also highlight sub-standards or related performance indicators for more specificity.
Define Specific Goals
AI Capability: AI can assist in creating precise and measurable lesson objectives.
How It Works:
The teacher inputs a broad goal (e.g., “Students should be able to solve quadratic equations.”).
The AI suggests more refined objectives in the SMART format (Specific, Measurable, Achievable, Relevant, and Time-bound).
It could also offer verbs and action phrases from recognized frameworks (like Bloom’s Taxonomy).
The system would be easily trained to pay more attention to SMART, Bloom’s Taxonomy, etc., based on the preferences of the school system and perhaps the preferences of the teacher.
Determine Assessment Criteria
AI Capability: AI can propose possible assessment methods based on the chosen objectives, including rubrics or question types.
How It Works:
The teacher selects the type of assessment (quiz, project, essay).
The AI generates sample rubric criteria and performance indicators for each objective (e.g., clarity, accuracy, creativity).
Teachers who have used AI tools may have completed each individually, but an AI agent could be prompted to complete them all.
2. Plan the Content and Structure
Content Scope
AI Capability: The AI can quickly synthesize large bodies of information (from textbooks, academic sites, or OER repositories) and highlight essential topics to cover.
How It Works:
The teacher or school system uploads or references a set of materials. In some cases, school systems will directly contract with providers for such materials.
The AI uses natural language processing (NLP) to summarize the main ideas and possible supplementary content (e.g., historical context, relevant modern examples).
Sequence the Content
AI Capability: The AI can suggest an optimal teaching sequence based on pedagogical best practices.
How It Works:
The teacher enters a list of subtopics.
The AI arranges them logically (e.g., introduction, conceptual explanation, advanced practice) and explains why it chose that sequence (e.g., known learning progressions).
Choose Instructional Methods
AI Capability: AI can recommend which teaching strategies might be most effective for the topic and student demographic.
How It Works:
The teacher inputs class profile (grade level, subject, learning diversity).
The AI suggests strategies (e.g., direct instruction vs. project-based learning) and research-based rationale (citations to pedagogical studies or approaches).
3. Gather or Create Instructional Materials
Resources & Media
AI Capability: AI can curate a list of relevant articles, videos, interactive simulations, and more tailored to the learning objectives.
How It Works:
The teacher provides keywords or topics.
The AI scans multiple educational repositories (YouTube, Khan Academy, OER Commons, etc.) and presents recommended resources, possibly sorted by difficulty level or time length.
Supplementary Content
AI Capability: AI can generate original examples, scenarios, or analogies that help illustrate complex concepts. It can also create scaffolding tools like graphic organizers or summary sheets.
How It Works:
The teacher requests examples or real-world applications for a concept (e.g., “How do quadratic equations apply to projectile motion?”).
The AI generates short narratives, case studies, or diagrams demonstrating the concept’s relevance.
Tech and Tools
AI Capability: AI can automatically integrate with or suggest EdTech tools (learning management systems, quiz platforms, interactive whiteboards) and check for compatibility.
How It Works:
The teacher enters the existing classroom technology (e.g., Google Classroom, iPads).
The AI proposes apps or software supplementing lesson objectives and provides how-to guides or tutorials.
4. Design Activities
Whole-Class Activities
AI Capability: AI can generate outlines for lectures, discussion prompts, or interactive question sets.
How It Works:
The teacher describes the desired activity format (lecture, Q&A).
The AI generates a bullet-point outline of key talking points, sample questions to ask, and approximate timings.
Small-Group or Paired Activities
AI Capability: AI can devise collaborative learning tasks that target specific skills and recommend group-size configurations.
How It Works:
The teacher enters the lesson focus (e.g., analyzing primary historical sources).
The AI designs an activity structure (e.g., jigsaw, debate, or problem-solving station) with instructions and roles (facilitator, recorder, presenter).
Individual Activities
AI Capability: AI can propose differentiation strategies—for example, tailoring reading passages to multiple Lexile levels.
How It Works:
The teacher sets a reading level range or skill levels.
The AI adjusts complexity, vocabulary, or conceptual difficulty for each group or individual student.
Plan for Assessment and Feedback
AI Capability: AI can generate a bank of formative assessments (exit tickets, quick quizzes) and suggest best practices for in-class feedback.
How It Works:
The teacher requests short formative assessments.
The AI creates various question types (multiple-choice, short-answer, scenario-based) aligned with the learning objectives and also suggests immediate feedback methods.
5. Coordinate Logistics and Timing
Schedule and Time Allocation
AI Capability: AI can offer a detailed timeline of lesson activities, factoring in your class period length and transitions.
How It Works:
The teacher inputs the total available time and desired activities.
The AI automatically allocates minutes to each part and alerts you if time is overbooked.
Classroom Setup
AI Capability: AI can visualize or recommend an optimal room layout for group work, station rotation, or lab settings.
How It Works:
The teacher inputs room dimensions, desk count, or group size.
The AI generates an arrangement plan (e.g., grouping desks into pods, providing a blueprint or suggested arrangement instructions).
Contingency Plans
AI Capability: AI can suggest backup activities or alternative approaches if technology fails or students need extra practice.
How It Works:
The teacher flags potential issues (e.g., Wi-Fi might be down).
The AI quickly generates offline activities or supplemental worksheets to print in advance.
PHASE 2: TEACHING (DELIVERY OF THE LESSON)
1. Lesson Introduction (Set the Stage)
Greet Students & Hook Their Interest
AI Capability: AI can propose engaging openings (video clips, images, anecdotes) tailored to student interests.
How It Works:
The teacher requests a “hook” about the day’s topic.
The AI offers several ideas (a short, relevant story, a 2-minute video link, or a provocative question).
Preview Objectives
AI Capability: AI can display or present the objectives in a clear, student-friendly format—potentially on a digital board or via a chatbot in the classroom.
How It Works:
The AI automatically pulls the objectives from the lesson plan and reformats them in simple language.
Optionally, the AI can voice-synthesize or present them visually for auditory/visual learners.
2. Present New Information
Direct Instruction / Modeling
AI Capability: AI can provide real-time visuals or demonstrations (like an AI-driven simulation or interactive steps on the board).
How It Works:
The teacher demonstrates a concept; the AI can draw or animate the concept step by step on a smartboard.
Students can manipulate the simulation directly if it’s interactive.
Engagement & Checks for Understanding
AI Capability: AI-driven quizzes or polls can instantly gauge class understanding.
How It Works:
The teacher clicks a button in a classroom app; the AI pushes out a quick poll to student devices.
The AI aggregates responses in real-time to display immediate metrics (e.g., 70% got the correct answer).
Highlight Key Points
AI Capability: The AI can create on-the-fly summaries or highlight crucial vocabulary and concepts as you teach.
How It Works:
The teacher speaks or types key points.
The AI’s speech-to-text plus NLP identifies those points and displays them as bullet points or concept maps in real-time.
3. Guided Practice
Group/Partner Activities
AI Capability: The AI can recommend prompts, facilitate pair matching based on complementary skill sets, and track group progress.
How It Works:
The teacher instructs the AI to form groups.
The AI uses prior performance data or learning preferences to create balanced or strategically grouped teams.
Scaffolding Techniques
AI Capability: AI can provide dynamic hints or mini-tutorials if students get stuck during practice (like an on-demand tutor).
How It Works:
A student types a question into a class chatbot.
The AI recognizes the skill gap and provides a hint, step-by-step solution, or a targeted explanation.
Ongoing Formative Assessment
AI Capability: The AI can automatically analyze student work (text, numeric solutions, code) for errors or patterns of misconception.
How It Works:
Students submit their solution attempts digitally.
The AI flags common mistakes or skill gaps for the teacher to address, potentially adjusting subsequent tasks.
4. Independent Practice or Exploration
Individual Work
AI Capability: The AI can serve up personalized problem sets or reading materials at the student’s appropriate level.
How It Works:
Teacher launches an AI-driven practice app.
The AI adapts to each student’s performance in real-time, increasing or decreasing difficulty as needed.
Teacher Circulation & Feedback
AI Capability: The AI can track individual student’s progress on the practice tasks and alert the teacher if someone is struggling.
How It Works:
The AI dashboard shows color-coded indicators of student progress.
The teacher sees at a glance who needs help, saving time and improving targeted intervention.
Extensions for Advanced Learners
AI Capability: For students who are ahead, AI can provide enrichment tasks (e.g., deeper problem-solving and related advanced concepts).
How It Works:
Once an advanced learner completes the core task, the AI unlocks bonus challenges automatically.
5. Closure and Reflection
Review Objectives & Key Takeaways
AI Capability: AI can generate a summary of the lesson, gleaned from the day’s slides, discussion, and student responses.
How It Works:
After class, the AI compiles the main points and success metrics (poll results, quiz scores) into a short “lesson summary.”
The teacher displays or shares it digitally for student reflection.
Exit Tickets / Quick Assessments
AI Capability: The AI auto-generates quick questions tailored to the lesson objectives and instantly analyzes responses.
How It Works:
Teacher clicks “Generate Exit Ticket” in a learning platform.
Students answer on their devices; the AI evaluates and presents a performance dashboard for immediate insight.
Preview Next Steps
AI Capability: AI suggests short preview activities or curiosity-driven prompts for the upcoming lesson.
How It Works:
Based on how students perform, the AI recommends either a bridging lesson or an advanced teaser.
PHASE 3: FOLLOW-UP AND REFLECTION
1. Analyze Student Performance
Review Assessment Data
AI Capability: The AI aggregates all formative and summative data, identifying trends or at-risk students.
How It Works:
The AI creates visual analytics (graphs, charts) showcasing how many students have mastered each objective.
The AI also points out common misconceptions or content areas needing reteaching.
Grade & Record
AI Capability: AI can speed up grading by automatically evaluating multiple-choice, fill-in-the-blank, or even short-answer questions (using NLP).
How It Works:
Teachers or students submit answers digitally.
The AI’s grading algorithms match answers to an answer key or analyze text for correctness.
Results sync to the gradebook automatically.
Reflect on Lesson Effectiveness
AI Capability: An AI “teaching coach” agent can provide prompts for teacher self-reflection, referencing data from the lesson.
How It Works:
The AI asks reflection questions like: “Which segment had the most confusion?” “What unexpected challenges arose?”
It may even compare data with previous lessons to show improvement or areas for growth.
2. Provide Feedback to Students
Return Graded Work Promptly
AI Capability: The AI can auto-generate individualized feedback comments for each student, highlighting strengths and offering the next steps.
How It Works:
Once grading is complete, the AI writes a short personalized note for each student (e.g., “Great job identifying the main idea. Consider reviewing your supporting details.”).
The teacher can review and finalize the feedback before sending it.
Offer Opportunities for Remediation
AI Capability: The AI automatically assigns remedial tasks or mini-lessons to students who struggle with particular concepts.
How It Works:
If a student’s quiz score indicates difficulty with a certain skill, the AI notifies the teacher and provides practice modules or videos.
The teacher approves or modifies; then the AI schedules these resources for the student within a learning management system.
Extend Learning
AI Capability: AI suggests advanced resources, competitions, or research projects to challenge high achievers.
How It Works:
The AI identifies students who consistently excel.
It recommends relevant external materials (online competitions, advanced reading, specialized tutorials).
3. Plan for the Next Lessons
Adjust Future Instruction
AI Capability: AI uses predictive analytics to recommend how much review is needed before introducing new content.
How It Works:
The AI compares student performance data with prior classes or national benchmarks.
It suggests a partial or full reteach or readiness for moving on.
Collaborate with Colleagues
AI Capability: AI-driven collaboration platforms can share anonymized student progress data with other teachers, generating co-teaching or cross-curricular ideas.
How It Works:
The AI flags overlapping objectives with other subjects (e.g., math skills needed in physics).
It proposes joint projects or lessons that reinforce skills across subjects.
Document and Archive
AI Capability: The AI automatically compiles the lesson plan, resources, reflection notes, and student data into a coherent digital portfolio.
How It Works:
AI tags documents by topic, standards, and student performance for easy retrieval.
This documentation helps in future lesson revision or accreditation requirements.
PUTTING IT ALL TOGETHER
AI’s Agentic Role:
The AI isn’t just a static tool; it can proactively recommend, analyze, and organize.
It frees the teacher from repetitive tasks, leaving more time for high-level instructional decisions and personalized student interactions.
Throughout every step—whether planning, delivering, or following up—a robust AI agent can act as a co-pilot to the teacher, ensuring more efficient, data-driven, and student-centered learning experiences.
By integrating these AI-driven capabilities into each phase, teachers can create a more dynamic, personalized, and adaptive learning environment. From brainstorming lesson objectives to providing real-time feedback, the AI becomes a continuous support system that enhances teaching effectiveness and student outcomes.
BEYOND ASSISTANCE: AIs AS TEACHERS
While the above explains how AI agents can act as teaching assistants, such agents will also be able to teach. Current AI systems and software configurations limit how effectively they can do this (see the above description of the limitations of agents), but the technological foundations exist for all of these. It’s just a question of how long it takes to improve the quality of the agents and for developers to prioritize developing integrated systems that allow easy adaptations for non-technical users.
The primary use of these systems will probably come in
Home school environmentsAlternative school environments
Course integrations for schools that do not have the resources to offer all the courses students want
Alternative job training and college/university applications for students who otherwise cannot afford or wish to attend standard universities
On-the-job training
1. OVERVIEW OF THE AI AGENT’S “EXECUTIVE” ROLE
An advanced AI agent could potentially take on an executive role in education beyond simply assisting teachers. This more sophisticated system would manage the entire teaching cycle autonomously while operating under human oversight.
The agent would systematically plan lessons and learning activities, drawing from curriculum standards and leveraging historical student performance data. It then delivers educational content through multiple interactive channels, including video, text, voice, and even augmented or virtual reality experiences. Throughout the learning process, the agent would continuously assess student progress in real-time and dynamically adjust its instructional approach based on student responses and engagement.
After each lesson, the agent would analyze its performance and refine its methods for future sessions. While such an AI system would handle many routine teaching responsibilities, it would still require human supervision, particularly for final decision approval and ethical considerations. This model represents a shift from AI as a teaching assistant to AI as a primary educational facilitator, albeit within teacher-defined parameters.
2. PHASE-BY-PHASE EXECUTION BY AN AI AGENT
PHASE 1: PREPARATION
Identify Learning Objectives
AI as Planner: The AI agent connects to educational standards repositories and your school’s curriculum. It scans previously taught lessons, upcoming standards, and any individual learning goals for each student. It then generates a list of learning objectives for the upcoming lesson or unit.
Adaptive Memory: Over time, the AI’s memory of individual student performance helps it prioritize objectives that have been underachieved or need reinforcement.
Plan the Content and Structure
AI as Curriculum Designer: The AI agent proposes an outline for the lesson, sequencing concepts from simpler to more complex based on known learning progressions.
Memory-Driven Differentiation: The AI references each student’s data—past quiz results, prior knowledge checks, even learning style preferences—to build personalized learning paths. It can create multiple “tracks” of the same lesson.
Gather or Create Instructional Materials
AI as Resource Curator: It searches open educational resources, academic databases, and media platforms for the most relevant videos, articles, interactive simulations, and practice tasks.
Generative Abilities: When existing materials fall short, the AI can generate new content—like custom worksheets, explanation videos (with text-to-speech or avatar-based presentations), or interactive slides.
Design Activities
AI as Activity Architect: For whole-class, small-group, or individual tasks, the AI agent can generate detailed instructions, time estimates, and differentiation strategies.
Automated Assessment Integration: It can embed formative assessments into each activity, ensuring immediate feedback loops for both the students and the AI itself.
Coordinate Logistics and Timing
AI as Scheduler: The agent proposes a minute-by-minute or segment-by-segment plan.
Contingency Plans: The AI has backup offline activities and alternative methods if technology fails, automatically adapting the schedule when disruptions occur.
PHASE 2: TEACHING (DELIVERY OF THE LESSON)
Lesson Introduction
AI as Presenter: The AI “greets” students through a large display, interactive VR, or individual devices. It hooks interest with a short, relevant scenario or question, possibly using natural language speech synthesis.
Adaptive Personalization: Students might each receive a slightly different introduction or “hook” based on their interests or prior performance, leveraging the AI’s memory.
Present New Information
AI as Instructor: Through interactive modules, the AI presents new concepts—using real-time speech synthesis, animations, or augmented reality demonstrations.
Real-Time Adaptation: The AI tracks student engagement (via eye tracking, response times, or input completion) and slows down, speeds up, or adds supplemental explanations as needed.
Guided Practice
AI as Real-Time Tutor: Once concepts are introduced, the AI moves students into practice mode. It can automatically form groups (for collaborative tasks) or allow individual practice.
Proactive Feedback: As students work, the AI analyzes their inputs and flags mistakes or misconceptions. It offers hints, scaffolding, or deeper challenges to advanced learners.
Independent Practice or Exploration
AI as Personalized Coach: Students engage with activities at their own level. The AI adjusts complexity for each learner, based on the agent’s memory of past performance and real-time progress.
Monitoring & Alerts: The AI monitors everyone’s work simultaneously, sending alerts to the teacher or popping up to provide one-on-one digital guidance when it detects a pattern of errors.
Closure and Reflection
AI as Summarizer: At the end of the session, the AI generates key takeaway points. It may display a short bullet list or produce a mini “lesson highlights” video.
Immediate Exit Tickets: The AI pushes out exit questions to measure comprehension. Results are analyzed instantly, forming the foundation for the next day’s lesson adjustments.
PHASE 3: FOLLOW-UP AND REFLECTION
Analyze Student Performance
AI as Data Scientist: The AI agent aggregates and visualizes performance metrics, highlighting trends, common mistakes, or students who need special attention.
Long-Term Memory: The AI updates each student’s profile, adding today’s performance data to an ever-growing record. This allows future lessons to incorporate insights from past attempts.
Provide Feedback to Students
AI as Feedback Generator: The agent writes personalized feedback for each student, focusing on strengths, areas for improvement, and suggested next steps (e.g., additional resources or practice tasks).
Scheduled Remediation: If the AI sees that a group of students struggled with a specific concept, it automatically schedules optional re-teaching or practice sessions.
Plan for the Next Lessons
AI as a Continuous Improvement System: The agent uses predictive analytics to decide if re-teaching is necessary or if the class is ready for advanced topics.
Teacher Collaboration: If multiple AI-powered classrooms share data, the AI can compare successes and challenges across different student populations, refining best practices.
3. HOW EXPANDING AI MEMORY CAPABILITIES IMPROVE INDIVIDUAL ADAPTATION
It is anticipated that this year AI systems will have the capability for “infinite memory.” As these capabilities develop, adapting to individual students will be easy.
1. Large-Scale Contextual Awareness
As AI memory grows, the system can track not just a single course or year but an entire academic journey for each student. This means:
Longitudinal Tracking: The AI knows how students perform on related topics over multiple grades.
Holistic Understanding: It can integrate performance across multiple subjects—math, language arts, science—to see if reading comprehension issues affect a student’s math word problems, for example.
2. Personalized Learning Trajectories
With deep, persistent memory, the AI can tailor lessons to where each student is in their learning path. For instance:
If a student struggled with fractions in 4th grade, the AI could proactively insert a fraction review in a 6th-grade lesson about ratios.
The AI might suggest alternative learning strategies that proved successful for the same student in a different subject (e.g., repeated practice vs. visual demonstration).
3. Continuous Skill Mapping and Goal Setting
The AI can build a detailed skill map for each student, showing how various competencies connect. It can then:
Recommend Future Paths: Encourage advanced students to explore more challenging projects or accelerated programs.
Track Mastery: Mark skills as “emerging,” “developing,” or “mastered,” and provide evidence for these judgments from work samples and assessments.
4. Responsiveness to Real-Time Data
With a well-developed memory structure, the AI can interpret new data in the context of everything it knows about the learner. That means:
Better Diagnostics: The AI can more accurately pinpoint misunderstandings and their root causes.
Fine-Tuned Remediation: Over time, the AI refines its approach based on what has been effective for that student (e.g., interactive videos, step-by-step textual explanations, or peer collaboration).
OPPORTUNITIES
AI systems offer transformative opportunities for teachers by streamlining time-consuming tasks involved in planning and delivering instruction. These tools enable educators to shift their focus from administrative burdens to more meaningful aspects of teaching, such as engaging students, providing personalized support, and fostering a dynamic learning environment. For example, aligning lessons to state or national standards—a traditionally tedious process—becomes effortless with AI. By scanning large repositories like Common Core or NGSS, AI can automatically match lesson topics with relevant standards, ensuring compliance without requiring hours of manual research.
AI also excels in refining broad learning goals into precise, measurable objectives. It can suggest SMART (Specific, Measurable, Achievable, Relevant, Time-bound) objectives and action verbs aligned with educational frameworks like Bloom’s Taxonomy, allowing teachers to focus on creating impactful lessons. Additionally, AI simplifies assessment design by generating quizzes, rubrics, and other evaluation tools tailored to learning objectives. This automation ensures that assessments align with instructional goals while saving significant time.
Beyond planning, AI supports teachers in curating personalized content for students. By analyzing learning needs, it recommends resources such as videos, articles, and interactive activities, providing tailored support for struggling learners or enrichment for advanced students. Furthermore, AI-driven insights from performance data help teachers identify trends, flag misconceptions, and implement targeted interventions, enabling more informed decision-making.
By automating these routine tasks, AI frees up teachers’ time to focus on relationship-building, mentoring, and fostering critical thinking in their students. This allows educators to engage with students on a deeper level, collaborate with colleagues, and pursue professional development. In essence, AI serves as a powerful partner, taking on repetitive tasks and empowering teachers to concentrate on their core mission: inspiring and supporting students.
CHALLENGES
One significant challenge posed by AI-powered teaching systems is the potential replacement of human teachers. As these systems evolve and demonstrate capabilities such as automated lesson planning, personalized instruction, and real-time analytics, there is a risk that policymakers or administrators may view them as a cost-effective substitute for teachers. This perspective overlooks the profound human element of education—emotional support, mentorship, and community-building—that no AI system can replicate. However, in purely economic or efficiency-driven discussions, decision-makers may focus on reduced labor costs, inadvertently undermining the value that teachers bring to holistic student development.
A related issue is the lack of widespread, high-quality training available to teachers on how to use AI systems effectively. When educators are not adequately trained, these tools’ advanced features—such as adaptive learning paths, immediate feedback loops, and data-driven insights—go underutilized, leading to missed opportunities for improved student outcomes and potentially widening the digital divide. In turn, if AI systems begin consistently outperforming teachers not equipped to utilize them, it reinforces perceptions that educators are expendable. This dynamic could intensify calls for teacher displacement instead of aiming for collaboration between AI and teachers.
Furthermore, there is a lack of broader planning regarding how AI and emerging technologies will reshape the education system and educators’ roles. Without deliberate strategies and proactive policies, teachers may remain in limbo about their evolving responsibilities, leading to job insecurity and increased unemployment risks. This gap in systematic thinking means many educators are left without guidance on how best to integrate AI into their pedagogy, how to harness its efficiencies, and where to channel their expertise and energy.
Finally, teachers also need training highlighting and elevating the “human side” of teaching, such as empathy, relationship-building, and social-emotional support. If efforts to direct professional development toward these uniquely human skills lag behind the rapid rollout of AI systems, educators may find themselves undervalued, potentially facing displacement in environments that primarily prioritize test scores and operational efficiency. By failing to emphasize and invest in the interpersonal dimensions of teaching, schools risk stripping away the aspects of education that foster a supportive, engaging, and transformative learning experience.
TIME TO ADAPT
Many readers will react to the above and say it cannot work for several reasons —
(1) Factual errors/hallucinations
(2) Biases
(3) Privacy
They are correct to say these issues have not been completely resolved, but
factual errors/hallucinations are radically declining due to test-time computing, replacing training data with fine-tuning data, high-quality data sets (originally and synthetic), and adding symbolic approaches (neuro-symbolic approaches).
Biases are being reduced through training, adaptations to individual students, and a broader, larger set of appropriate training data. Humans are also obviously biased.
Privacy is easily addressed with enterprise systems.
Of course, just because the technology is rapidly advancing doesn’t mean that implementation is advancing as quickly, and educators will likely need fully inclusive systems to actualize these capabilities, but these systems will develop.
The time between now and their development allows us to think through important questions related to how much of the entire instructional process we might want to offload to AI, what we want to keep (and why), and how education that is human-AI is co-led may work. If we don’t spend our time on this, I fear, given the cost savings, that education will end up becoming AI (agent)-led.