Leave this slide up while people come in, so they can open the slides and the parking lot at pd.sidarous.net/ac/ before you start. In the HTML version, each item jumps to its slide. The links use slide numbers (#4 is slide 4), so if you add or remove slides, update them. A 45-minute session for the 5 to 12 teachers who picked it from the session list. The goal: everyone leaves with a way to decide when AI fits an accommodation, one support they made for their own class, and prompts they can use again. Core message: AI is good at the time-consuming part of accommodations: rewriting, chunking, reformatting, and making supports. The IEP or 504 team decides what each student gets. The teacher checks everything before a student sees it. Before the session, email the group: - Bring a laptop and a lesson or handout you'll use soon. - Think of the accommodations you see most in your classes. - Sign in to Gemini with your CPS account. - In Google Docs, turn on Tools → Preferences → Enable Markdown (it's off by default, and the demo uses it). - The slides and the parking lot are at https://pd.sidarous.net/ac/. (The full draft is in accommodations-pd-email.md.) Teaching sequence: welcome (2 min), my journey (5), "Would you use AI here?" think-pair-share and the flowchart (11), Top 5 and the live demo (11), work time and sharing (13), wrap-up (3). If anything runs long, shorten the demo: the slides cover what you skip. Materials, next to this deck: the planning notes (accommodations-pd-notes.md), a draft of the participant Google Doc (accommodations-pd-google-doc.md: prompts for the Top 5, the flowchart, what not to do, Google Docs tips, Gem instructions, more ideas), and a draft of the email (accommodations-pd-email.md). The flowchart is built by accommodations-pd-flowchart.py. Sources: the CPS AI Guidebook (Version 6.0) and the Instructional GenAI Playbook (Version 1.0), both converted to Markdown in this folder, and Google's Docs Editors Help for the Markdown features. The demo uses Draw Your Game Character, a Game Programming 1 handout. Privacy: the two students in my journey are described by their needs only. Keep names, initials, and other identifying details out of the slides and out of what you say.

The session overview, the email to send beforehand, and the sources are in the notes on the Contents slide.

Half the session is the room talking or making something: the think-pair-share, work time, and sharing add up to 24 of the 45 minutes. Say that up front so people know this isn't a lecture. There's no buffer. If the think-pair-share runs long, cut demo slides, not work time.

Lead with humility. Some people in the room use AI every day, some are skeptical, and some have real ethical objections. All of those are reasonable, and nobody has to change their mind today. But we can't spend 45 minutes on AI fundamentals, so today is about one narrow use: helping carry out accommodations students already have. The parking lot is a publicly editable Google Doc, linked from pd.sidarous.net/ac/ with the slides. When a bigger question comes up, say "great question, let's park it," and ask the person to add it (or add it yourself). Follow up after the session, then actually follow up. Since anyone with the link can edit it, remind people: questions only, and no student names or details.

This is the whole session in three lines. Come back to it when a scenario or a demo moment gets confusing. AI never decides which accommodations a student gets: that's an individualized team decision based on data. AI helps carry out what the plan already says, and the teacher checks every output before a student sees it. The human is in the loop at both ends.

About 5 minutes for this section. The point is honesty: what didn't work matters as much as what did, and it sets the tone for a room with different views on AI.

Describe the need, not the students: no names, no initials, no details that could identify them. That's the same rule teachers should follow when they describe a student's need to AI. Java is unforgiving: one missing semicolon or a capital letter in the wrong place, and the program won't run. For students who struggle to read the directions at all, the directions themselves became the barrier.

My first instinct was to explain more: longer, more detailed instructions. When the text is already hard, more of it didn't help much. What worked was the opposite: cutting questions that repeated each other, lowering the text complexity, and adding more screenshots. I made some of the screenshots and AI made some. Land the line at the bottom: more words isn't more support. It's also why "reading-level rewrite" is in the Top 5 later.

One support was being able to see a labeled grid and put shapes on it. In Processing, y goes down from the top, which confuses everyone at first, and a low reader can't easily keep that straight from text alone. The grid shows it. Drawing a grid with every 20 pixels labeled is simple but tedious: exactly the kind of work AI is good at. And it's a clear accommodation, not a modification: students still have to work out every coordinate themselves. The grid just makes the canvas visible. The grid is used in the Draw Your Game Character, Board Game in Progress, and Code Detective handouts.

The first output wasn't the final one. The grid first came out with only the big gridlines and cut off at the edges. The next version wasn't cut off, but it still had only the big lines. I kept asking for changes until it was right: larger, nothing cut off, every 20 pixels labeled. Be clear that the first two pictures are re-creations: I didn't save the in-between versions, so these were made afterward to show what they looked like. The point for work time: your first output won't be your final one either. Asking for changes is normal, not a sign you did it wrong. That's the human in the loop.

About 11 minutes for this section: 8 for the think-pair-share, 3 for the flowchart. This replaces a lecture on accommodations vs. modifications and on what not to do. Teachers work it out themselves, and the flowchart at the end names what they just did.

Think first, so everyone commits to an answer before the loudest voices set the direction. With 5 to 12 people, everyone can share in the whole group. The situations are on the next slide. Leave it up for the whole think and pair time.

The four are a clear no (A), a clear yes (B), a no that surprises people (C), and an it-depends (D). With that mix, the room doesn't hear "AI is bad" or "AI is great." It hears "it depends, and here's how to decide." Swap for D if you like: a plan lists word banks, and you ask Gemini for a word-bank version of your vocabulary quiz. (It depends: on a vocabulary quiz, a word bank may give away the answers. Check the plan's wording, or ask the case manager.) Other situations, if a group finishes early: translating an IEP meeting notice for a family (no: official communication needs a qualified translator), or having AI grade essays when a plan says not to count off for spelling (careful: tell AI what to ignore, and you make the final call).

These items use * so they appear one per click in the HTML version. Keep them as *, not -. Reveal each one after that group has shared, not before. A: The main problem: deciding which accommodations a student gets is an individualized team decision, so this is a no even with no student data at all. On the data: CPS does allow student data, including identifying information and information from a student's record, in Gemini signed in to your CPS account (Guidebook, Guidance for Educators and Staff; the Playbook: "PII can only be entered in approved products with data share agreements"). Never in unapproved tools or a personal account. My personal conviction is to keep student PII out of AI entirely, and I'd ask you to anonymize as much as you can: describe the need, not the student. B: Guided notes are in the plan, they're tedious to make by hand, and you can check them in a minute against your own slides. Ask for blanks on the key terms only, so students still process the content. One twist worth mentioning if there's time: if the plan says "copy of notes" instead, fill-in-the-blank notes may not meet it, because a copy means the full notes. Read the plan's exact wording. (This replaced the grid example, since the grid was just in my journey.) C: If the plan says "read aloud by an adult," text-to-speech may not meet it, even if you think it's just as good. Deliver it as written, and bring other ideas to the case manager or team. D: This is where accommodations vs. modifications comes in. In a science class, a lower-reading-level version of a reading can be an accommodation: the goal is the science. In an ELA lesson on the author's word choice, the original words are the whole point, so a rewrite changes what the student is learning. The CPS Playbook asks teachers to keep "the culturally and linguistically responsive complex texts and tasks central to the lesson at the intended level of rigor," and to make sure "scaling down" a text for readability "does not strip away the complex concepts or critical thinking required by the grade-level standard." Better supports here: read-aloud of the original, a vocabulary preview, or guided questions. If the plan itself includes a modified text, that's the team's decision to follow. Say this out loud: a modification isn't wrong. Some students, especially those with the biggest needs, have modifications in their IEP. It's only a problem when a teacher, or an AI, makes one the plan doesn't include. The CPS Playbook makes the same distinction: end-of-unit assessments are given without modification "except per IEPs and/or language proficiency levels."

Walk the four situations through it quickly. A stops at question 1 (it's not in a plan yet: deciding is the team's job). B goes all the way down. C stops at question 2. D stops at question 3: a rewrite changes what students learn in a lesson about the author's words. Question 3 is accommodations vs. modifications: an accommodation changes how a student gets to the learning or shows it; a modification changes what they're expected to learn. "Simplify this" prompts can quietly cross that line, which the demo will show. Question 4 covers tools and privacy. Use Gemini, the District-approved chatbot (CPS AI Guidebook, Version 6.0), signed in with your CPS account, and no other tools. CPS allows student data there, including identifying information. My personal conviction is to keep student PII out of it, so I'd ask you to anonymize as much as you can: describe the need, not the student. Question 5: AI is good at the tedious making: rewriting, reformatting, chunking, making organizers and checklists. It's not the one to make decisions. The flowchart will also be in the participant Google Doc.

About 11 minutes: 3 for the Top 5, 8 for the demo, ending with the what-not-to-do slide.

All five are tedious to make by hand, quick to check, and common in plans. Each one should keep the learning target the same. 1. Reading-level rewrite: keep the key vocabulary, bolded and defined, and keep every requirement. 2. Graphic organizer: for a reading or a topic. 3. Fewer items, same skills: ask AI to choose 10 of 20 problems that still cover every skill, and to explain its choices. If the 10 it drops are all the hard multi-step ones, that's a modification. 4. Guided notes: blanks for key terms only, so students still process the content. 5. Break up a project: covers planning and organizing; supports chunking and extended time. The demo shows 1 and 5. More ideas are on the backup slide at the end and in the Google Doc.

The handout is Draw Your Game Character (Game Programming 1, Unit 2). In Task 3, students plan a game character on paper before they code it. It has the hardest text of the Processing assignments so far, and it's where the planning grid comes in. It's planning on graph paper, not code, so any teacher can judge whether a rewrite keeps the rigor. The key idea in the task: a rectangle is placed by its top-left corner, but an ellipse is placed by its center. Watch for it. Live demo, in one Gemini chat, signed in with your CPS account: 1. Paste Task 3 (from the Google Doc, with Edit → Copy as Markdown so the lists stay lists). The robot picture won't come along. 2. Send prompt 1 (the rewrite) and check it against the next slide. 3. Send prompt 2 (checkpoints) and look at the reading level. 4. Send the follow-up, then paste into Google Docs and turn the steps into checkboxes. The slides after this one show what happened when I tried it on October 29, as a backup if the live demo goes wrong. The robot is inspired by P5, the robot in Getting Started with Processing by Casey Reas and Ben Fry.

This is the original Task 3 from my Google Doc, the version I pasted into Gemini (it includes the Google Drawing option I added for students who'd rather draw digitally). The full original is in assets/accommodations-pd-task3-original-top.png and assets/accommodations-pd-task3-original-bottom.png. Point to the outlined sentence before you send any prompts, so the room is watching for it. In my test, Gemini kept all three label steps and dropped this sentence in both the rewrite and the checkpoints.

The full prompt I used: "I teach a high school computer science class. Rewrite the assignment directions below at a 5th-grade reading level for students who need text at a lower reading level. Keep every requirement. Don't add or remove anything students have to do. Keep these terms. Bold each one and define it in a few simple words the first time it appears: canvas, pixel, rectangle, ellipse, top-left corner, center, decomposition. Use short sentences and one action per bullet." Point out what the prompt does: it names exactly what changes (reading level, sentence length) and what stays (every requirement, the key terms). Never just "make this easier." One prompt per step, all in the same chat: each output gets checked on its own, and each prompt can become a saved prompt or a Gem.

Ask the room to check along with you. These are every requirement in Task 3, plus its key idea. In my test, the rewrite passed every requirement check: the grid, the labels going down, 8 shapes, 3 colors, the labels for each shape, and the drawing order. All the key terms were bolded and defined. But two things slipped through, on the next two slides.

Being inspired by a game and copying from it are different things, and the line right above says "Make your own original design." It's a one-word change that's easy to miss, which is exactly why a person reviews every version.

Both label steps are still there, so the requirement check passes. But the one sentence that explains the difference is gone, and that difference is what students need to understand (the handout's question 7 asks about it directly). Why it happened: the prompt said "keep every requirement," and Gemini took that literally. It kept every to-do and dropped the explanation, because the explanation isn't a to-do. It did the same thing again in the checkpoints version. The fix: name the key idea in your prompt ("Keep this idea: a rectangle is placed by its corner, an ellipse by its center"), or type it back in yourself. This is a quiet way a "simpler" version can lower the rigor: the steps survive, the understanding doesn't.

This is "break up a project" from the Top 5. It uses the original text so mistakes in the rewrite don't carry over. Notice what this prompt doesn't say: the reading level. Watch what happens.

"Establish the identity," "strictly following the shape, color, and boundary constraints": this is harder than the original. Gemini doesn't carry an accommodation from one prompt to the next unless you say it again. This is the best argument for Gems: put "5th-grade reading level, keep every requirement and key idea" in a Gem's instructions, and it's on for every prompt. (The HOWTO: Gemini Gems deck in this folder shows how to build one.) The good news: every checkbox here is an action, and the credit line kept its original meaning.

The dollar signs are LaTeX, a math format. Google Docs' Paste from Markdown doesn't convert it, so it shows up as raw code. Ask for plain text, like (x, y). "Parameters" adds a word the task never asked students to learn (the original says "arguments"). For a low reader, every new term costs something.

The follow-up brought the reading level back down, put one requirement on each checkbox, and fixed the coordinates. But the corner vs. center idea was still missing, the second time Gemini dropped it. The line I'd type under Checkpoint 4: "A rectangle starts at its top-left corner. An ellipse starts at its center. Look at the robot's head and eyes." Human in the loop doesn't always mean another prompt. Sometimes it's faster to just type the sentence. In my Google Doc, I also added things Gemini can't: the planning grid as a Google Drawing students can draw on, and the labeled robot picture (pictures don't come along when you copy text into Gemini, so put them back).

Gemini writes its answers in Markdown: bold, headings, lists, tables. The copy button under an answer copies it as Markdown. Paste from Markdown in Google Docs turns that into real formatting, instead of stray ** and # marks. Two gotchas: - A plain Ctrl+V doesn't convert. Use the right-click menu or Edit → Paste from Markdown. - Copy as Markdown and Paste from Markdown are off by default. Turn them on once with Tools → Preferences → Enable Markdown. The setting is per person and sticks. Going the other way, Edit → Copy as Markdown copies part of a Doc (like a handout section) into Gemini with its lists and tables kept. And File → Download → Markdown (.md) exports a whole Doc. Check tables after pasting, since some people report they don't always convert cleanly. Sources: Google Workspace Updates, "Import and export Markdown in Google Docs" (July 2024), https://workspaceupdates.googleblog.com/2024/07/import-and-export-markdown-in-google-docs.html; Google Docs Editors Help, "Use Markdown in Google Docs, Slides, & Drawings," https://support.google.com/docs/answer/12014036. Menu names change, so check them again before the session.

After pasting the checkpoints, right-click any bullet and choose the checkbox option to turn the whole list into a checklist. Students then click a box to check off a step, and it gets crossed out. This is "break up a project" made usable: students can see what's done and what's next. This screenshot is from my October 29 test, before the follow-up prompt. Notice that some of these checkboxes are facts, not steps ("Each box is 20 pixels"). A student can check those off without doing anything. That's one more thing to catch when you review: every checkbox should be something the student does.

The first three came up in the scenarios. The last two are what the room just watched: Gemini kept every to-do but dropped the key idea twice, and changed "inspired by" to "copy." On the last one: you can ask AI to list every requirement and show where it kept each one. That makes your check faster, but it isn't a replacement for it. AI can claim it kept something it actually dropped. You're the check. Privacy details: CPS allows student data, including identifying information and information from a student's record, in Gemini signed in to your CPS account, because Gemini has an enterprise data share agreement with the District (Guidebook; the Playbook: "PII can only be entered in approved products with data share agreements"). Never put student data in unapproved tools or a personal account: the Playbook says they may train on it. Say this plainly: my personal conviction is to keep student PII out of AI entirely. You're allowed to enter it in Gemini, but I'd ask you to anonymize as much as possible: describe the need, not the student. Laws that apply include FERPA, COPPA, and Illinois's SOPPA. More don'ts for the Google Doc: don't ask AI to "make this easier" (say exactly what changes and what stays); don't make one "IEP version" for everyone; don't label materials "easy version"; don't use AI detectors (they're more likely to flag students using speech-to-text or assistive tools, and English learners); don't write progress notes from memory; and where it fits the plan, teach students to use supports themselves.

About 13 minutes: 7 to make, 6 to share.

Teachers can work on their own and share, or work with a partner. Seven minutes is enough to paste, prompt, and get a first output, maybe one round of changes. Most people won't finish, and that's fine: the Google Doc has the prompts so they can pick it back up. Walk the room. Common snags: not signed in to the CPS account, Paste from Markdown not turned on, or a prompt that doesn't name the reading level or the key ideas.

With 5 to 12 people, that's 30 seconds to a minute each, or less if partners share together. "What did you have to fix?" is the more important question. It shows the room that everyone's first output needed a person, which is the point of the session.

Add the link to the Google Doc here once it exists. The slides and the parking lot (a publicly editable Google Doc) stay up at pd.sidarous.net/ac/ after the session. Gems: a saved prompt or a Gem is the reusable part. You state the accommodation once (reading level, keep every requirement and key idea) instead of in every prompt, which fixes what went wrong in the demo. The HOWTO: Gemini Gems deck in this folder walks through building one. Then read the takeaway on the last slide.

The takeaway. Leave this up while people pack up, and thank them for coming.

Backup slide for questions about other uses. These are also in the Google Doc. Chunk directions: numbered steps, one action each, with no requirements dropped. Sentence starters and word banks: frames and vocabulary that fit the task (a word bank on a vocabulary quiz may be a modification). "I can" rubrics: a rubric rewritten as self-check statements. Worked examples: a solved problem with each step labeled. Plain-language feedback: your comments, or a confusing error message, in short sentences. Vocabulary preview: key terms on one page before the lesson. Choice board: two or three ways to show the same skill. Clean up a worksheet: fewer items per page, more white space. Starting-a-task checklist: a short routine students mark off. Summary before reading: 3 to 5 sentences before a long reading. Practice quiz with explanations: each answer explains why it's right or wrong. Audio: text-to-speech of the exact text is free and needs no AI (Edge's Read aloud, for example). AI can prepare text for text-to-speech by spelling out symbols, describing diagrams, and writing code the way you'd say it. Podcast-style AI audio (Gemini's Audio Overview) is a new text that needs checking, and it may be a modification if students listen instead of reading the real text.