15 Implementation Practicalities

Plan for the tool to fail. The pedagogy should work even when the technology does not.

If AI integration only works for some students, it is not integration. It is exclusion with better technology.

15.1 When Things Go Wrong

AI tools are powerful but not infallible. When implementing AI-enhanced teaching, technical issues are inevitable. The question is not whether something will fail, but whether you have a plan when it does.

15.1.1 The Essential Preparations

Have backup platforms. If your primary tool goes down, students should be able to switch to an alternative. Design assignments that work with any AI tool — ChatGPT, Claude, Gemini, Copilot — rather than depending on one. Specify this in your assignment instructions.

Build in deadline flexibility. Include a clear policy: if the AI platform is unavailable during the assignment period, students can request an extension. This removes anxiety and prevents a wave of panicked emails.

Prepare offline alternatives. For any AI-enhanced activity, have a non-AI version ready. A conversation simulation can fall back to a written case analysis. An interactive exercise can fall back to a paper worksheet. The learning objectives should be achievable either way.

Test before you launch. Run every prompt yourself before assigning it to students. Check that the AI stays in character, produces useful output, and handles edge cases. A prompt that works for you in a calm office may fail when 80 students use slightly different wording.

15.1.2 Common Issues and Quick Fixes

Problem	Response
Platform is down	Switch to alternative tool. Extend deadline if needed.
AI not staying in character	Start a fresh conversation. Check full prompt was copied.
AI giving inconsistent quality	Try a different model or platform. Simplify the task.
Student cannot access tool	Offer office hours to run it together. Provide alternative.
Student says “is this cheating?”	“This IS the assignment. You’re graded on critical use, not avoidance.”
Conversation gets too long and degrades	Start a new conversation. Restate key context explicitly.

15.1.3 Risk Assessment

Risk	Probability	Impact	Mitigation
Service outage	Medium	High	Multiple platforms, offline alternatives
Quality degradation	High	Medium	Benchmark testing, platform switching
Data loss	Medium	High	Local backups, screenshot important outputs
Security/privacy incident	Low	Critical	Enterprise tools, data sanitisation

15.2 Accessibility and Inclusion

AI integration creates new opportunities but also new barriers. If you design only for students with reliable internet, modern devices, and strong English skills, you are excluding a significant portion of your cohort.

15.2.1 The Digital Divide

Not all students have equal access to technology. Before assigning AI-enhanced work, consider:

Device availability. Some students access coursework only through phones or shared computers. Design activities that work on mobile browsers, not just desktops.
Internet reliability. Rural and low-income students may have poor connectivity. Allow activities to be completed in short sessions rather than requiring extended online time. Provide campus lab time as an option.
Cost barriers. Free-tier tools have usage limits. Design assignments around freely available platforms. If your institution provides enterprise AI access, make sure students know how to use it.
Digital literacy. Not all students arrive with the same comfort level. A five-minute demo of “here is where you paste the prompt” removes a barrier that feels obvious to you but is not obvious to everyone.

15.2.2 Universal Design Principles

Design AI activities that work for the widest range of students:

Multiple means of engagement. Offer text-based, voice-based, and visual interaction options where possible. Allow students to pace interactions at their own speed.

Multiple means of expression. Let students demonstrate learning in different ways — written reflection, verbal explanation, visual presentation, or process documentation. A student who struggles with written English may demonstrate excellent critical thinking verbally.

Multiple means of representation. Request AI output in different formats — text summaries, bullet points, step-by-step breakdowns — so students can choose what works for their learning style.

15.2.3 Specific Accommodations

Students with visual impairments. Prioritise text-based AI tools with strong screen reader support. Ask AI to provide structured, hierarchical responses rather than visual content. Avoid prompts that depend on “see above” or visual references.

Students with hearing impairments. Ensure all AI interactions are available in text. If using voice-based AI features, provide text alternatives.

Neurodivergent students. AI conversations can be overwhelming. Allow breaks, provide templates that structure the interaction, and keep prompts focused on one task at a time. The “one prompt, one job” principle from CRAFT is especially important here.

Non-native English speakers. AI tools are primarily trained on English. Allow students to work in their preferred language where possible. Include cultural context in prompts to avoid Western-centric examples.

15.2.4 The Process Assessment Safeguard

The process-over-product assessment approach described in the Assessment chapter has a built-in accessibility benefit: the process component should not require AI use. A student who keeps a research journal, annotates readings, or documents their decision-making in any form should be able to submit equivalent process evidence. The assessment measures engaged thinking, not AI use. This ensures students with limited access, disability-related barriers, or personal preferences against AI can demonstrate the same learning through alternative means.

15.2.5 Course Design Checklist

Before launching any AI-enhanced activity:

Are chosen AI tools accessible on mobile and desktop?
Can students participate using free-tier tools?
Are alternative (non-AI) pathways available?
Is campus lab time available for students without reliable home access?
Are instructions clear enough for students with no prior AI experience?
Are timing accommodations built in for students who need extended time?
Does the activity work for students using screen readers or assistive technology?
Are examples culturally diverse rather than Western-centric?

15.3 Your Action Step

Pick one AI-enhanced activity you are planning. Run through the checklist above. Identify the two biggest accessibility gaps and fix them before launching. Then ask yourself: if the AI platform went down the night before, what would students do? If you do not have a good answer, create the backup plan now.

# Implementation Practicalities > Plan for the tool to fail. The pedagogy should work even when the technology does not. > If AI integration only works for some students, it is not integration. It is exclusion with better technology. ## When Things Go Wrong AI tools are powerful but not infallible. When implementing AI-enhanced teaching, technical issues are inevitable. The question is not whether something will fail, but whether you have a plan when it does. ### The Essential Preparations **Have backup platforms.** If your primary tool goes down, students should be able to switch to an alternative. Design assignments that work with any AI tool — ChatGPT, Claude, Gemini, Copilot — rather than depending on one. Specify this in your assignment instructions. **Build in deadline flexibility.** Include a clear policy: if the AI platform is unavailable during the assignment period, students can request an extension. This removes anxiety and prevents a wave of panicked emails. **Prepare offline alternatives.** For any AI-enhanced activity, have a non-AI version ready. A conversation simulation can fall back to a written case analysis. An interactive exercise can fall back to a paper worksheet. The learning objectives should be achievable either way. **Test before you launch.** Run every prompt yourself before assigning it to students. Check that the AI stays in character, produces useful output, and handles edge cases. A prompt that works for you in a calm office may fail when 80 students use slightly different wording. ### Common Issues and Quick Fixes | Problem | Response | |---|---| | Platform is down | Switch to alternative tool. Extend deadline if needed. | | AI not staying in character | Start a fresh conversation. Check full prompt was copied. | | AI giving inconsistent quality | Try a different model or platform. Simplify the task. | | Student cannot access tool | Offer office hours to run it together. Provide alternative. | | Student says "is this cheating?" | "This IS the assignment. You're graded on critical use, not avoidance." | | Conversation gets too long and degrades | Start a new conversation. Restate key context explicitly. | ### Risk Assessment | Risk | Probability | Impact | Mitigation | |---|---|---|---| | Service outage | Medium | High | Multiple platforms, offline alternatives | | Quality degradation | High | Medium | Benchmark testing, platform switching | | Data loss | Medium | High | Local backups, screenshot important outputs | | Security/privacy incident | Low | Critical | Enterprise tools, data sanitisation | ## Accessibility and Inclusion AI integration creates new opportunities but also new barriers. If you design only for students with reliable internet, modern devices, and strong English skills, you are excluding a significant portion of your cohort. ### The Digital Divide Not all students have equal access to technology. Before assigning AI-enhanced work, consider: - **Device availability.** Some students access coursework only through phones or shared computers. Design activities that work on mobile browsers, not just desktops. - **Internet reliability.** Rural and low-income students may have poor connectivity. Allow activities to be completed in short sessions rather than requiring extended online time. Provide campus lab time as an option. - **Cost barriers.** Free-tier tools have usage limits. Design assignments around freely available platforms. If your institution provides enterprise AI access, make sure students know how to use it. - **Digital literacy.** Not all students arrive with the same comfort level. A five-minute demo of "here is where you paste the prompt" removes a barrier that feels obvious to you but is not obvious to everyone. ### Universal Design Principles Design AI activities that work for the widest range of students: **Multiple means of engagement.** Offer text-based, voice-based, and visual interaction options where possible. Allow students to pace interactions at their own speed. **Multiple means of expression.** Let students demonstrate learning in different ways — written reflection, verbal explanation, visual presentation, or process documentation. A student who struggles with written English may demonstrate excellent critical thinking verbally. **Multiple means of representation.** Request AI output in different formats — text summaries, bullet points, step-by-step breakdowns — so students can choose what works for their learning style. ### Specific Accommodations **Students with visual impairments.** Prioritise text-based AI tools with strong screen reader support. Ask AI to provide structured, hierarchical responses rather than visual content. Avoid prompts that depend on "see above" or visual references. **Students with hearing impairments.** Ensure all AI interactions are available in text. If using voice-based AI features, provide text alternatives. **Neurodivergent students.** AI conversations can be overwhelming. Allow breaks, provide templates that structure the interaction, and keep prompts focused on one task at a time. The "one prompt, one job" principle from CRAFT is especially important here. **Non-native English speakers.** AI tools are primarily trained on English. Allow students to work in their preferred language where possible. Include cultural context in prompts to avoid Western-centric examples. ### The Process Assessment Safeguard The process-over-product assessment approach described in the Assessment chapter has a built-in accessibility benefit: the process component should not require AI use. A student who keeps a research journal, annotates readings, or documents their decision-making in any form should be able to submit equivalent process evidence. The assessment measures engaged thinking, not AI use. This ensures students with limited access, disability-related barriers, or personal preferences against AI can demonstrate the same learning through alternative means. ### Course Design Checklist Before launching any AI-enhanced activity: - [ ] Are chosen AI tools accessible on mobile and desktop? - [ ] Can students participate using free-tier tools? - [ ] Are alternative (non-AI) pathways available? - [ ] Is campus lab time available for students without reliable home access? - [ ] Are instructions clear enough for students with no prior AI experience? - [ ] Are timing accommodations built in for students who need extended time? - [ ] Does the activity work for students using screen readers or assistive technology? - [ ] Are examples culturally diverse rather than Western-centric? ## Your Action Step Pick one AI-enhanced activity you are planning. Run through the checklist above. Identify the two biggest accessibility gaps and fix them before launching. Then ask yourself: if the AI platform went down the night before, what would students do? If you do not have a good answer, create the backup plan now.