Twelve Wednesdays, One Movement: Reflecting on Ctrl + AI + Teach
By Leandra McGriff, Education Outreach Lead, Infosys Foundation USA | May 21, 2026
When we kicked off Ctrl + AI + Teach, I knew I wanted to build something different. Not another keynote. Not another panel where teachers nod politely and leave with a slide deck they will never open again. I wanted twelve real Wednesday nights — free, hands-on, screen-shared — where teachers from every corner of the country could log in, try the tool live with me, and actually walk away with something they could use first thing Monday morning.
Twelve sessions later, I am still a little overwhelmed in the best way.
I got to spend an hour each week with elementary teachers from rural districts, high school CS leads, K–8 specialists, ESL coaches, and curriculum directors. Some of you joined every single session. Some of you found us on Session 7 because a colleague forwarded the link. A few of you stayed in the chat after the recording stopped just to ask one more question. All of it mattered.
This post is part recap, part thank-you, and part roadmap for any educator who wants to catch up on the recordings or run a session in their own building.
What Ctrl + AI + Teach was built to do
The series was twelve free PD webinars hosted by Infosys Foundation USA, built around a few core beliefs:
- AI literacy is not a CS-teacher problem. It belongs to every educator, K–12.
- Teachers learn best when they get to actually click the buttons, not just watch someone else click them.
- Free tools are non-negotiable. Equity dies the moment a session requires a paid subscription.
- The goal is not to make teachers into prompt engineers. It is to make them confident enough to bring AI into their classrooms without flinching.
Every session was recorded, every session had a downloadable resource, and every session was open to anyone with an internet connection.
Session 1 — AI 101 for Teachers: Demystifying the Black Box
We started where every honest conversation about AI has to start: what it actually is, and what it absolutely is not. I introduced the CTFC framework — Context, Task, Format, Constraints — using the mnemonic "cats tackle flying chickens," because if you cannot remember it, you will not use it. We sorted AI into reactive (spell check), predictive (Khan Academy, Turnitin), and generative (ChatGPT, Gemini, Claude), and I made the case that generative AI requires the most teacher judgment, not the least.
The opener was a one-word icebreaker: type your first reaction to AI. Curiosity. Scared. Suspicious. I told the room what I tell myself: that is not a bug, that is a feature. Skepticism is a posture worth keeping. Ammani asked a sharp question about copyright liability when AI reproduces someone else's work, and the conversation that followed made the whole session feel less like a webinar and more like a faculty room with the good coffee.
Session 2 — Prompt Power-UP: Unlock AI's Potential
Session 2 went deeper into prompting. We layered four advanced techniques on top of CTFC: role-based prompts ("you are a merchant in colonial Boston in 1775"), step-by-step instructions, progressive scaffolding, and few-shot learning where you feed the AI a sample of your own writing so the output sounds like you. We modeled differentiation across three reading levels, cognitively tiered assessments, and the parent emails that always seem to need writing at 9 PM.
I also gave my honest tool recommendations: Claude for math and CS, Gemini for language-heavy work, ChatGPT for quick classroom activities, and Perplexity for fact-checking. The moment I will not forget came during the Prompt Makeover Challenge, when an attendee submitted: "I am clickbait — explain how you trick people into clicking." Exactly the kind of unexpected angle that makes a class lean in.
Session 3 — AI Adventure Time: Sparking Curiosity in Young Learners
This is the one that made me grin the whole way through. Paige Best, the CSTA 2023 CS Teaching Excellence Award winner with twenty-five years of classroom experience, came in with a 147-slide Day of AI library and a clear answer to every elementary teacher who has ever asked, "But how do I do this with first graders?"
Paige walked us through an Artificial vs. Natural sorting game, Google Quick Draw, an AI Perception handout where students apply rules (fur, whiskers, four legs, smaller than a laundry basket) and watch the model misfire, and the picture book Duck Rabbit as a starting point for human vs. AI perception. When the model got something wrong, she said the line I have been quoting ever since: "This is wonderful. This is how you have those conversations." And on the teachers who are still hesitant: "If we don't teach our students to understand it, that's when things get misused." Pre-K through fifth grade teachers — go watch this one twice.
Session 4 — AI High School Hacking
Session 4 was for the high school crowd. We built a movie poster genre classifier across three skill levels in the same hour: a no-code version using Google Teachable Machine, a block-based version pairing Machine Learning for Kids with Scratch 3, and a text-based version where students complete nine guided to-dos inside a JavaScript and ML5.js web app I had pre-built with Claude. Five downloadable files went out after the session — the training app, the testing app, two student versions, and a teacher guide.
When Teachable Machine misclassified a family movie poster as action, I called it a gift. It is hard to teach students about training data, sampling, and bias when the model behaves perfectly. The wrong answers are the lesson.
Session 5 — From Overwhelmed to Empowered
By Session 5, our regulars were ready for something more demanding. I introduced the CRAFT framework — Context, Role, Audience, Format, Task — as a refinement of CTFC, easier to remember and more flexible. We worked three live tiers in one hour. Beginner: a differentiated water cycle worksheet in ChatGPT. Intermediate: an iterative chemistry lab built across multiple turns in Gemini. Advanced: a 30-question American Revolution test bank generated and compared across ChatGPT, Gemini, and Claude side by side.
The night's comic relief came when I asked Claude to auto-fill a parent newsletter template and it cheerfully invented a student named Ella Johnson, fabricated a class celebration, and generated a fake teacher contact. Perfect teachable moment about why your prompts have to do the specifying — the AI will fill any gap you leave for it, and not always in your favor.
Session 6 — Code the Future: Claude and JavaScript Integration
This was the session I was most nervous about, because we live-coded. The first half stayed in Claude's browser interface, where natural language alone produced three single-file HTML tools: an ecosystem simulator with sliders for plants, prey, and predators; a fractions visualizer using a CSS grid bar model; and a short story idea generator with dropdowns for character, setting, and conflict. Anyone without Claude access just used Gemini or ChatGPT and followed along.
The second half moved to Claude Code in the terminal. I wrote the JavaScript scaffold for a vocabulary matching memory game called Quiz Quest myself — the card data array, the deck-building logic — then handed Claude Code the structured prompt and watched it ask refinement questions about animation style, timer duration, feedback type, and storage. By the end of the hour we had a finished classroom-ready game with 3D card flips, a streak counter, confetti animation, and an editable vocab manager. James, who was managing chat, told me later that the energy in the comments shifted the moment that game ran live.
Session 7 — AI for All: Enhancing Accessibility and Inclusivity
I started this session with something I do not always volunteer: I am hard of hearing. No hearing in my left ear. I had recently come out of a surgery that I hoped might restore some of it. I told the room because I wanted them to know I was not teaching accessibility from the outside.
We worked through five real classroom barriers, each paired with a free tool. Language and belonging — ChatGPT generating a bilingual water cycle text with English science terms bolded inside the Spanish version, and a third-grade reading-level adaptation of the same content. Reading and visual access — the Helper Bird browser extension running on a National Geographic article, with immersive reading, color-coded parts of speech, dyslexia fonts, and picture-based word definitions. Auditory processing — 11 Labs speech-to-text for live transcription, free for educators on request. Writing and physical expression — Google Docs voice typing with a clear rule for students: dictation is for drafting, your brain is for thinking, your eyes are for editing. Executive function — a ChatGPT prompt that breaks a five-paragraph essay assignment into a numbered checklist with one explicit instruction: do not write the essay for them.
I told the room about a former student named Marcus, who had motor and speech difficulties but used basic sign language to crack jokes and ask the smartest questions in class. Removing physical barriers does not reduce cognitive challenge. It just lets the cognitive challenge be the part students struggle with, which is how it should have been all along.
For me, captions are not a nice feature on Netflix. They are a lifeline. AI accessibility tools should not be either, but for now they are some of the best ones we have, and they are free.
Session 8 — AI for Visuals: Image and Video Creation Made Easy
Session 8 ran a little rough on my end — technical difficulties knocked out some of the visual overlays in the opening minutes — and the audience was patient with me, which I have not stopped being grateful for. We covered five sections: explaining hard concepts, student-facing materials, accessibility visuals, teacher prep, and video.
Gemini ImageFX (which I kept calling Nano Banana, sorry not sorry) generated a flat 3rd-grade water cycle diagram and a labeled, semi-realistic 7th-grade version from the same base prompt — a beautiful illustration of how prompt vocabulary alone differentiates instruction. ChatGPT produced a cinematic photorealistic version of a mysterious door storybook scene next to Nano Banana's whimsical version, so teachers could see how style descriptors change the emotional register of an image. We compared a butterfly life cycle infographic across Nano Banana and ChatGPT, and the chat consistently favored ChatGPT for text accuracy and inclusive imagery.
The video segment used Sora and gave us a perfect teachable error: the second medieval marketplace clip showed a character grabbing at empty air. I paused, asked the audience to spot the mistake, and we had a real conversation about what AI does not know. AI has never lived in the real world. It has never seen what we have seen.
Session 9 — Scratch and AI Spark Creative Coding
Session 9 was for CS teachers, especially the ones working with younger students. We built a movie-recommendation chatbot using Machine Learning for Kids and Scratch 3 — labeling training data into three categories (action, romance, comedy), entering the kinds of phrases a student might actually type, training the model, and connecting it to a Scratch chatbot that returns movie recommendations with confidence scores.
We added an if/else block so the chatbot answers "I am unsure" when its confidence drops below a threshold instead of forcing a guess. When the chatbot returned the comedy Up in response to the word "scary," I told the room: this is not a failure, this is sparse training data, and that is exactly what your students need to see. The bias does not come from the machine. It comes from the human who chose what to feed it.
I opened by asking attendees to rate, on a scale of 1 to 10, how magical their students think AI is. Linda's answer was the one I keep coming back to: "When it comes to images and videos, nine. When it comes to using AI as the new Google, two." Our students are sharper than we sometimes give them credit for.
Session 10 — PBL with AI
Project-based learning with AI works under one rule: laptops stay closed for the first 15 to 20 minutes. Students build foundational knowledge before AI enters the room. I demonstrated three live projects:
- Elementary — Instruction Engineers. Students hunt their school for vague vs. clear posted instructions, sort them on green and yellow sticky notes, redesign the vague ones into precise step-by-step versions on posters, and then stress-test their engineered instructions in ChatGPT. Garbage in, garbage out, made tactile.
- Middle School — Historical Hallucination Hunt. Students build a 10-fact wall on a historical event using only physical sources — textbooks, library books, encyclopedias — with citations on every index card. Then I prompted ChatGPT to generate a Boston Tea Party summary with three subtle errors planted on purpose. Wrong year. Wrong number of tea chests. Students had to catch them.
- High School — Persona Prototype. Students create a four-section character deep dive (textual evidence, symbols, relationships, motivations and contradictions) for a literary character — I used Juliet — then prompt ChatGPT with a detailed in-character roleplay and write a two-to-three-page critique of where the AI captured the character and where it failed.
When I asked which AI summary of the Boston Tea Party was more valuable for learning — the helpful one or the tricky one — Anna and Floene both picked the tricky one. That is the right answer. When AI tries to trick you, you have to become an expert to catch it.
Session 11 — Middle School Mastery
Middle school teachers got three projects designed to scale from sixth to eighth grade.
The Source Citation Sprint required no computers at all. Students received a short AI-generated paragraph — I used a sample about the James Webb Space Telescope that included the fabricated claim that it had "already proven there is life on exoplanets" — and had to underline checkable facts, circle broad claims, verify against credible sources like NASA, and mark anything unverifiable with a reason. Confident-sounding text is not the same as confirmed information. That sentence belongs on a poster in every middle school classroom.
The AI Art Spec Sheet asked students to write a one-page avatar specification before generating any image — audience, style, hair, skin tone, accessories, lighting, setting, constraints — with no real photos permitted. When I typed the vague prompt "create a cartoon version of me" into both ChatGPT and Claude, both immediately asked me to upload a photo. Linda jumped into the chat: "No photos!" That moment was the lesson. Privacy is not a school safety measure. It is a habit students need to take with them everywhere they ever use AI.
The Human vs. AI Reasoning Lab had students evaluate the same prompt — "explain why the seasons change" — across ChatGPT outputs with different constraints, scoring for accuracy, clarity, and audience targeting, and watching for false-balance language like "scientists agree" and "some people think."
Session 12 — Python Powerhouse
We closed with Python, and I opened with a sentence I meant: Python is not the hard part. Vulnerability is. The fear of looking dumb in front of your students keeps more teachers out of CS than any syntax error ever has.
The first half was for Python beginners. Day 1 should cover four concepts and four only: variables, input, output, and conditionals. A variable is a sticky note on a lunchbox. A conditional is the same logic a streaming algorithm uses to decide what to show next. Break the code in front of your students on purpose. Take out a colon. Trigger the syntax error. Then walk them through three steps: read the full error message, go to the line number Python identifies, and ask "what did I forget?" instead of "what did I do wrong?" Use Google Colab, Replit, or Trinket so nothing has to be installed.
The second half was for Python-experienced teachers. I built a live news classifier in Google Colab — credible, questionable, and satire — using Machine Learning for Kids and Python's requests and csv libraries. The code sent headlines to the trained model, received JSON predictions with confidence scores, bucketed them as high confidence, needs review, or low confidence, logged incorrect predictions to a mistakes.csv file, and ran a four-question human verifier check that students could grade themselves.
Live debugging is its own kind of vulnerable. The audience caught my indentation errors, my misplaced brackets, and a stray F-string character in my URL. I will say what I said in the moment: thank you. The pressure is real, but errors are how you teach.
What I will carry with me
The wins I will hold onto from this series are not the slide decks or the resource files. They are the moments in chat. Linda calling out the photo upload. Ammani asking the copyright question. Anna and Floene picking the AI that tries to trick you. Paige Best telling a hesitant teacher why we cannot afford to opt out of teaching this.
Twelve weeks of meeting teachers where they are — from the kindergarten teacher who had never opened ChatGPT before Session 1 to the AP Computer Science teacher who showed up in Session 12 and started debugging my Python alongside me — reminded me of something I already knew but needed to feel again: teachers are the most adaptive professional class in this country. Drop a new technology in their lap and they will figure out how to make it useful for kids, fast, and they will do it whether or not anyone gives them paid time to learn it.
What we owe them in return is exactly what Ctrl + AI + Teach was designed to be. Free. Practical. Honest about what AI can do and equally honest about what it cannot. Built around real classrooms, real students, and real Tuesdays at 11:42 PM when the worksheets still are not done.
If you missed a session, every recording lives on the Infosys Foundation USA YouTube channel. Pick the one closest to your grade level or your subject and start there. Bring a colleague. Try one thing in class on Monday. Then come tell me how it went — that part of this series was always my favorite.
Thank you for showing up. All twelve weeks of it.
