AI-Powered
Differentiation
A portrait of learning from a national crew of educators exploring how AI can differentiate instruction for every learner — while keeping the teacher as the architect of what good learning looks like.
How the crew engaged
Each dot is one session. Paired = synchronous 20–30 min peer conversation. Solo = individual 10–15 min guided reflection.
Success Signs are observable evidence indicators from the AI for Deeper Learning (AI for DL) framework — mapped from this crew's Small Wins and session transcripts.
High draft volume reflects active private reflection — not low engagement. Drafted Ideas and Small Wins are only visible to the individual learner who received or created them.
Each node is one crew member. Lines show paired sessions; line weight reflects number of sessions together. 24 cross-organizational connections formed over the crew period.
The shape of the learning
The crew signed up to get better at one concrete thing: using AI to differentiate for a wide range of learners. What they walked away with was a clearer sense of who has to own that work. AI is fast and never gets tired, but it doesn't actually know what good differentiation looks like. That judgment sits with the teacher.
You can hear the shift across the 38 sessions and 78 documented Ideas and Small Wins. Early on, people described AI as a way to save time. By the presentations of learning, they were describing a thought partner they steer. The educators who got the most out of it treated differentiation as something they design. They set the vision, chose the entry points, and decided how hard the work should be, then handed the repetitive precision to the tool.
Two ideas kept coming back. The first was that equity means access to rigor. For this crew, differentiation was never about making the work easier. It was about building a real way into complex, grade-level thinking for every student, including those with the most significant needs. The second was about time. When AI absorbs the labor-intensive prep, it gives hours back, and the crew was clear about where those hours should go: relationships, feedback, and the kind of struggle that learning still depends on.
One pattern is worth a facilitator's attention. Almost all of this crew's richest thinking stayed in private Draft rather than reaching the group. The opening for next time is to help people carry that private experimentation into shared practice, and to build in the individual-student focus that several secondary teachers named as their own next step.
AI doesn't necessarily know what good differentiation looks like.
You have to be the one architecting the breadth and scope of what is happening.
AI removes barriers, even on standardized tests and especially those with severe cognitive disabilities.
Five themes that surfaced
What came up repeatedly and with specificity across transcripts, Ideas, and Small Wins — not just once, and not in passing.
AI is good at staying precise and it never tires, but it has no sense of what good differentiation actually looks like. That vision belongs to the teacher, who sets the scope and the level of rigor and then puts AI to work carrying it out. The tool takes some of the load off; the judgment stays human.
For this crew, differentiation was never about making the work easier. It was about giving every student a genuine way into complex, grade-level thinking. They watched for AI's habit of over-scaffolding and "generating toward the center," and pushed back by building multiple entry points into hard work rather than around it.
The clearest example came from special education. A crew member used AI to adapt a state standardized alternate assessment for students with significant cognitive disabilities, keeping the standard intact while making it reachable. What normally takes hours of hand-building was ready in about a minute.
A lot of what eats a teacher's week is prep: building differentiated tasks, writing IEP paperwork, leveling texts. When AI takes on that work, the hours come back. Crew members were clear that the payoff is human. It means more time with kids when they need it, and in one case, actually getting home at a reasonable hour.
This was the crew's sharpest caution. AI is built to take friction away, but you can't really learn without some of it. The harder skill is calibrating how much is useful, so students still have to think while AI prompts the struggle rather than doing the work for them.
Competency Strand Activity
This map shows where this differentiation crew's learning clustered across the five competency strands of the AI for Deeper Learning framework. Bold items reflect areas with documented evidence in session transcripts and Launched Small Wins. Lighter items represent areas not yet surfaced in this crew's data.
Bar indicators reflect depth of evidence in crew data — not performance quality or evaluative grade. Bold items = documented evidence present. Lighter items = not yet surfaced in this crew's data.
sideby's AI for Deeper Learning Framework is informed by the UNESCO AI Competency Framework for Teachers, the colleague.ai AICE Framework, Teach AI with EC + OECD & code.org frameworks for AI learning, and the Hewlett Foundation Deeper Learning Competencies — evolving as AI evolves and shifts what this technology can mean for deeper learning and education.
Where evidence is strongest
Five Learning Targets (LTs) for the AI-Powered Differentiation badge, grounded in the four course activities — from Foundations & Goal-Setting to the Presentation of Learning. Evidence depth reflects richness of crew data — not a performance grade or evaluation.
Learning Targets map to the AI-Powered Differentiation (AIPD) badge and the five strands of sideby's AI for Deeper Learning framework (PH · MA · CT · AL · LM). Evidence depth reflects the richness of documented crew data — session transcripts and Launched Small Wins — not a performance grade.
Differentiation moves in action
Concrete differentiation practices crew members put to work with real students and staff — each using AI to execute a vision the educator architected. These are documented steps drawn from session transcripts, not aspirations. Roles are de-identified.
Uploaded a state standardized alternate assessment into Colleague AI and asked it to keep alignment with the standard while modifying the content for accessibility, so it worked for students with significant cognitive disabilities. What would normally be hours of hand-building came back as a finished, individualized packet almost immediately, and the standard stayed intact.
Over three years, built and trained purpose-specific GPTs for a caseload of sixteen students, working from color-coded learner profiles and interests like anime and gaming. Using AI this way changed how the job gets done and gave back hours that used to disappear into evening prep.
Designed a game where a Depth-of-Knowledge nudge shifts the challenge level, so every student lands on work that is rigorous for them. Also runs a daily custom GEM co-teacher, opening each morning with "today's date is May the 4th. What are we doing today?" and layering in GLAD and SIOP strategies for English-language development.
Used Brisk to rewrite core texts across about twenty reading levels and built Gemini choice boards for enrichment and keyboarding practice. This crew member was firm about the limit of the tools: "You can't replace the human aspect of teaching."
Built a transformers-themed lesson around a student's interests and set up a Copilot IEP agent to move through the paperwork faster. The point was what the saved time bought: "less documenting more hanging out with kids when they need it."
Designed adult professional learning with choice boards and multiple breakout paths, so staff could start wherever they were, whether skeptics or power users. Worked within a board resolution on student-facing generative AI and AI policy, keeping the work moving inside real constraints.
Used Copilot to generate sight-word stories built around a student's favorite characters, then had the learner draw their answers as a comprehension check. For a young, emergent reader, that opened up more than one way to show understanding.
Coaching principal interns across the whole range of AI adoption, met each person where they were. For some that meant setting a calendar time zone; for others it meant building GEMs. Throughout, AI stayed a thought partner rather than a doer. Differentiation applies to adults, too.
What got documented
Small Wins are things crew members tried and learned from. Ideas are helpful thoughts, concepts, or goals. Both are generated through the sideby platform — some reflected back automatically from session transcripts, some written directly by learners. Drafted = private to learner; Launched = shared with the full crew.
A crew member used Colleague AI to adapt a state standardized alternate assessment for students with significant cognitive disabilities, prompting it to hold alignment with the assessment's standards and protocols while modifying the content for accessibility. Work that would normally be extremely time-consuming and labor-intensive became an individualized assessment packet in under a minute, without compromising the integrity of the standard. The takeaway they named: AI can remove barriers even on standardized assessments, especially for students with the most significant support needs.
A crew member reflected on the teacher's role in architecting learning experiences and using AI to make precise, repeated cognitive work possible. Their throughline: AI can help produce high-quality differentiation, but teachers remain responsible for giving students multiple entry points and opportunities to synthesize and demonstrate academic skills.
AI doesn't know what good differentiation looks like; the human does. Keep the teacher as the one deciding the vision and the level of rigor, and let AI stay precise and consistent in service of it.
Learning requires some struggle, and AI is built to remove it. The design task is calibrating the right amount of friction, using AI as a thought partner that prompts thinking rather than a doer that shortcuts it.
Student passions like transformers, favorite characters, anime, and gaming are legitimate levers for rigor. AI makes it fast to build interest-based tasks that still reach grade-level thinking.
When AI absorbs the labor-intensive prep and paperwork, protect the reclaimed hours for the human work: less documenting, more time with students, and getting home at a reasonable hour.
Crew voice across the experience
Direct quotes from sessions and Small Wins from Crew members who included teachers, coaches, case managers, and directors of professional learning.
Equitable learning demands equitable access to rigorous, complex thinking and work.
What AI can do is stay consistently precise.
So less documenting more hanging out with kids when they need it.
How important it is to keep humans in the work with it as people grow more confident in it.
If I have to choose between meeting the needs of my students and documenting it, documenting it has never won.
It felt like taking it off the page and giving me some time with the words there.
You can't replace the human aspect of teaching.
A lot of times we need different and out-of-the-box thinking, and AI isn't that.
AI can easily differentiate with a given set of instructions to fit different learners' needs, allowing the individualization and access for so much more than ever before.
And the teaching is as important as any of the AI technology that I'm going to be using, and frankly, a little bit more so.
Teachers are still responsible for giving students multiple entry points and opportunities.
Platform & course terminology
What comes next for this crew
Three time horizons based on what's in the data — for thinking about follow-up, iteration, or a future cohort.
- Surface the two Launched Small Wins in a crew-wide discussion — 97% of captured learning stayed in private Draft
- Invite crew members to Launch one differentiation move each, with the design principle underneath, not just the action
- Name the 24 unique cross-org pairings — that network is a resource beyond the badge
- Prompt each crew member to identify one concrete next step before the crew disperses
- LT 5 (evaluating impact) needs more structure — consider a mid-badge checkpoint to capture success signs and iteration
- Secondary teachers named an individual-student focus as their next step — build that into the sequence explicitly
- Document the daily GEM / IEP-agent co-teacher pattern as a transferable move for the next cohort
- Adult-learner differentiation surfaced organically — make "differentiate for your staff" an explicit design prompt
- Most of the richest thinking stayed in Draft. What would make Launching feel safer or more useful?
- Learning still needs friction, and AI is built to remove it. How do we help teachers engineer the right amount?
- When a board policy restricts student-facing generative AI, what does responsible differentiation look like inside that constraint?
- How do we keep the teacher as architect — humans in the loop — as the tools get faster and more capable?
The teacher is still the architect.
This crew used AI to remove barriers and hand time back — then spent that time on the human work: connection, feedback, and the productive struggle that learning still requires. AI stayed precise; the educators held the vision of what good differentiation looks like.