# Pain points in innovating in education with AI ## How I read this request I focused on **transcripts where education, training, or academic governance is explicit**, including: - IITM / TDS / OPPE / viva discussions - NIE / Srikanth Nadhamuni discussions - IIM Bangalore discussions - Ashoka University discussions - Straive's internal training discussions that mirror educational constraints - SWAYAM / IITM RFP discussions Also, not every blocker is literally the **Board** or **Senate** saying "no". In these transcripts, innovation is often stopped by the same function through different mechanisms: - Senate statutes - committee calendars and approval chains - management skepticism - IT / security policy - support and evaluation staffing - messy operational data - faculty politics and legitimacy concerns ## Executive summary The recurring pain points are: | Pain point | How it gets stopped | Representative transcripts | | -------------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | | Leaders are not yet convinced AI is real enough to redesign around | Management committees first demand persuasion | `2026-03-11 Srikanth Nadhamuni.md`, `2026-03-12-nie-ai-roadmap/transcript.md` | | Even supportive institutions move slowly | PGP chairs, HoDs, workshop slots, and committee calendars delay action | `2026-02-19 Sourav Anand IIMB.md`, `2026-02-13 Shantanu Bhanja Ashoka.md` | | Assessment redesign hits hard statutory boundaries | Senate statutes constrain grading structure | `2026-01-28 TDS.md` | | Nobody wants to reopen formal curriculum approval | Senate / DCC escalation is seen as too costly politically | `2026-01-07 TDS.md`, `2026-01-14 Palani IITM.md` | | AI-enabled coursework needs tool freedom institutions often do not allow | IT/security/procurement policies block installs, GitHub, extensions, public repos | `2025-09-09 Internal Gen AI Capability Building Rajat.md` | | Integrity is harder, not easier, in the AI era | Students can use ChatGPT, proxy helpers, and organized external markets | `2025-11-03 IITM bot and viva.md`, `2025-08-18 IITM LLM Viva demo.md` | | Human support remains the hidden bottleneck | Evaluation, TA support, and exception handling do not scale automatically | `2025-09-09 Internal Gen AI Capability Building Rajat.md`, `2025-11-03 IITM bot and viva.md` | | Institutional data is too messy to automate cleanly | Manual patching and multiple source-of-truths break dashboards and policy decisions | `2025-11-20 Jaidev.md` | | Faculty capability is not the only issue; incentives and legitimacy matter | Tools alone do not change behavior; institutions want sanctioned, durable formats | `2026-03-11 Srikanth Nadhamuni.md`, `2026-02-13 Shantanu Bhanja Ashoka.md` | | Student readiness and UX can defeat good ideas | Cognitive overload, unclear specs, and cluttered interfaces create failure | `2026-03-11 IITM BS Python OPPE.md` | | Faculty also worry AI may arrive too early or weaken learning habits | "Calculator before tables" anxiety and well-being concerns slow adoption | `2026-03-12-nie-ai-roadmap/transcript.md` | | Procurement processes are poorly matched to AI | Long-term RFPs and static contracts lag model change and trust requirements | `2025-12-18 AI for Swayam IITM RFP.md` | | Underneath it all is employability panic | Institutions know work is changing faster than curricula | `2026-03-12-nie-ai-roadmap/transcript.md`, `2026-02-07 Vishnu.md` | ## Detailed pain points ### 1. Senior leaders first have to be convinced that AI is real, urgent, and not just hype This came up most clearly in the NIE discussions. Before curriculum redesign, workshops, or AI-enabled pedagogy, the first hurdle is **management belief**. **How it gets stopped** - Management committees ask whether AI is hype. - Innovators have to spend energy creating urgency before they can discuss implementation. - The first audience is not students or faculty, but skeptical decision-makers. **Quotes** > "Some of them are sort of old guard. They're asking, you know... 'I mean, is this all hype?' and all that stuff."\ > — `2026-03-11 Srikanth Nadhamuni.md` > "The first part is more... less for faculty, more for management committee because these guys make decisions."\ > — `2026-03-11 Srikanth Nadhamuni.md` > "A couple of questions came up at the management committee. We have this group asking, 'Is AI hype? Is there reality to AI or is it just hype?'"\ > — `2026-03-12-nie-ai-roadmap/transcript.md` **Interpretation** The pain point is not lack of ideas. It is that innovators must first win a **credibility battle** with senior leadership before they are allowed to run the experiment. ### 2. Even when people like the idea, committee calendars and institutional slotting slow everything down At IIM Bangalore and Ashoka, the resistance is often softer. People are interested, but the innovation gets delayed by **orientation schedules, chair approvals, faculty calendars, and political sequencing**. **How it gets stopped** - No slot in orientation - Waiting for a chair or senior faculty member - Need to meet someone in person because email stalls - Workshops treated as "somewhere there" rather than urgent **Quotes** > "I had sent it to the PGP chair... She has not been able to come back because I think the orientation schedule is pretty tight."\ > — `2026-02-19 Sourav Anand IIMB.md` > "The one with Shubo... I had sent him a message and we had also talked, and I was not very clear about his response. And so I had to go and meet him because I thought that over email it might not happen."\ > — `2026-02-19 Sourav Anand IIMB.md` > "Suppose I want to do this at IIM Bangalore. Won't there be a huge resistance from people saying, 'Hey, what are you doing?'"\ > — `2026-02-19 Sourav Anand IIMB.md` **Interpretation** The blocker is not always principled opposition. Often it is **institutional drag**: there is no obvious owner who can say yes quickly. ### 3. Formal Senate rules constrain assessment redesign even when everyone agrees the current format is weak This is one of the clearest Board/Senate-style blockers in the transcripts. **How it gets stopped** - Staff want to change weights or redesign assessment - Senate statutes define minimum weights for proctored components - The conversation shifts from pedagogy to compliance **Quotes** > "We can't do that unfortunately because for a proctored exam, minimum of 20% is required as per Senate statutes."\ > — `2026-01-28 TDS.md` > "The students got Project 1, Project 2, which they lost hope [on] and we said End-term will have something to ease up at the end of the course."\ > — `2026-01-28 TDS.md` > "We need to find ways to ask meaningful questions."\ > — `2026-01-28 TDS.md` **Interpretation** The pain point is not only "what should we test?" but "what are we legally allowed to count?" AI-era assessment innovation runs into **pre-AI governance assumptions**. ### 4. Nobody wants to reopen full curriculum approval unless absolutely necessary This is a different kind of Senate blocker: not a rule, but **approval fatigue**. **How it gets stopped** - Credit changes, syllabus changes, and structural changes imply Senate / DCC movement - Teams avoid the formal process unless they absolutely must - Innovation gets forced into the existing credit/time box **Quotes** > "That needs to go up to the Senate. Sorry, we have no appetite to take stuff to the Senate anymore."\ > — `2026-01-07 TDS.md` > "Do we also have a constraint that we cannot go beyond 8 weeks?"\ > — `2026-01-07 TDS.md` **Interpretation** A lot of AI innovation therefore happens as **workarounds inside the existing shell** rather than through clean structural redesign. The `2026-01-14 Palani IITM.md` discussion reinforces the same pattern: once course structure and delivery format are entangled, teams first negotiate within the current setup rather than reopen the official structure. ### 5. Delivery and platform rules create "minimum acceptable" operational gates before pedagogy can even start This surfaced in the NPTEL/IITM discussions. **How it gets stopped** - Recording formats and delivery constraints must satisfy platform owners - Even when platforms are interested, there are minimum operational requirements - Remote/hybrid delivery is negotiated case by case rather than designed as default **Quotes** > "They are really hungry for recording. So anytime you say that we are ready to do this, they are very happy."\ > — `2026-01-14 Palani IITM.md` > "Maybe because he is in Singapore, they may allow [remote recording]. And it is possible. But sir, if you are here, we would expect you to at least record some minimal amount."\ > — `2026-01-14 Palani IITM.md` > "One is lecture hours is not added, going by the traditional metric."\ > — `2026-01-29 Saji IITM DOMS NPTEL.md` > "The narrative form of writing and evaluation is a tough process in the NPTEL platform. They haven't developed that."\ > — `2026-01-29 Saji IITM DOMS NPTEL.md` **Interpretation** This is a softer governance blocker: platforms are willing, but **innovation still has to fit delivery rules**. Even an AI-native course must negotiate studio, format, and operational acceptability. ### 6. Security and IT policy can make AI-enabled education impossible even when the content already exists This is one of the strongest themes in the Rajat / Straive training discussions, and it generalizes directly to universities. **How it gets stopped** - VS Code installs are blocked - extensions are blocked - GitHub is blocked - public repos trigger policy concerns - LMS hosting does not solve the real problem because the course requires action, not just reading **Quotes** > "The evaluation ended up being too much work... that was one bottleneck."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "Second bottleneck is half, or more than half, the sites that people need to access for that course were blocked."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "The problem is not that the content is not accessible. The problem is that the assessments that I'm telling them to do are not allowed."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "Airbus has a very strong data policy. They say, 'Oh, you know what, you can't be using any of these LLMs.'"\ > — `2025-10-29 IITM Palani.md` > "We cannot expand GenAI capability without GenAI capability. We cannot expand GenAI capability with GenAI constraints. So it's a chicken and egg problem."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "I have exhausted my karma. You want to do it, you reach out."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` **Interpretation** This is a very important Academic Council lesson: **AI education is not just a curriculum problem**. It is also a permissions, procurement, and sandbox-design problem. ### 7. Human evaluation and support remain the hidden cost AI may generate content cheaply, but support, review, and exception handling remain human-heavy. **How it gets stopped** - no one is staffed to handle support - evaluation still needs human triage - pilots are easy; scaled operations are not **Quotes** > "At IIT, I have a bunch of teaching assistants who are doing the evaluation... but still there is some work. And that was not something that we had set up within Straive."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "At IIT Madras, yes, there are a set of teaching assistants... In short, yes, we need somebody to support."\ > — `2025-09-09 Internal Gen AI Capability Building Rajat.md` > "I don't have a TA, and I will have to grade all of them."\ > — `2025-10-29 IITM Palani.md` > "So, Level 1, this time we are doing for about 1,000... 2,000 people... there are 20 people who are sitting and just doing this, ensuring that these people are genuine."\ > — `2025-11-03 IITM bot and viva.md` **Interpretation** This is why many AI education pilots look magical in demos and painful in real operations. **Support load** becomes the real scaling limit. ### 8. Academic integrity is now an arms race against both ChatGPT and organized cheating markets The IITM viva transcripts are explicit that the problem is no longer hypothetical. **How it gets stopped** - students use live ChatGPT during viva - identity cannot be established by email alone - external coaching/proxy markets form around known rubrics and proctors - staff are forced into industrial-scale filtering **Quotes** > "I have to validate who the person answering is, not just the email ID. I could give it to 10 of my friends..."\ > — `2025-11-03 IITM bot and viva.md` > "Usually ChatGPT... they'll paste this question, and they'll read out the answer."\ > — `2025-11-03 IITM bot and viva.md` > "Everybody's getting WhatsApp messages, they're getting mails saying, 'We'll help you, we'll help you.' They even, to the extent, know who the proctor is going to be."\ > — `2025-11-03 IITM bot and viva.md` > "So 2,100 we're doing a Level 1..."\ > — `2025-11-03 IITM bot and viva.md` **Interpretation** The pain point is not merely "cheating exists." It is that AI has made cheating **cheap, individualized, and scalable**, forcing institutions to redesign assessment around proof of understanding rather than hope. ### 9. Broken source-of-truth data makes academic automation brittle and politically risky The Jaidev discussion is one of the clearest descriptions of why academic AI systems often fail for reasons that look "non-AI". **How it gets stopped** - manual patches override official records - multiple sources have to be reconciled - dashboards are politically sensitive because numbers drive Senate-facing decisions - automation surfaces broken processes rather than fixing them **Quotes** > "There is a lot of manual patching involved of the database."\ > — `2025-11-20 Jaidev.md` > "BigQuery is not necessarily the source of truth for everything. You have to compile data for a given student from multiple places..."\ > — `2025-11-20 Jaidev.md` > "If that has to stop, it means that some students will have to suffer."\ > — `2025-11-20 Jaidev.md` > "The problem is that the numbers are never 100% accurate because they have to be manually compiled through multiple sources."\ > — `2025-11-20 Jaidev.md` **Interpretation** This is a board/senate issue in disguise. If the numbers going into dashboards, progression rules, and transcripts are unstable, leadership becomes understandably conservative about AI automation. ### 10. Faculty incentives and mental models are often a bigger blocker than tool availability The NIE transcripts are especially strong on this. **How it gets stopped** - faculty are not yet thinking in AI-native ways - leadership expects speed gains to automatically translate into output - institutions optimize for rankings and grants, but behavior change requires more than tools **Quotes** > "I need them to think out of the box. They are not thinking like this yet."\ > — `2026-03-11 Srikanth Nadhamuni.md` > "NIE is not doing too well... we are not producing enough research."\ > — `2026-03-11 Srikanth Nadhamuni.md` > "One of the key things we figured out was... how do we get NIE into the top 100... we're not producing enough research."\ > — `2026-03-11 Srikanth Nadhamuni.md` **Interpretation** This is not a pure capability problem. It is an **incentive and legitimacy** problem: faculty must feel that AI changes what they are rewarded for, not just the speed of existing work. ### 11. Institutions want legitimacy, documentation, and permanence — not just flashy one-off workshops Ashoka surfaced this very clearly. **How it gets stopped** - a one-off workshop is seen as insufficiently serious - people want a series, a document, and something that can become institutionally real - innovators want speed; institutions want legitimacy **Quotes** > "It should not be just a one-off workshop."\ > — `2026-02-13 Shantanu Bhanja Ashoka.md` > "We can do a series of these workshops and create a kind of a document out of this whole idea."\ > — `2026-02-13 Shantanu Bhanja Ashoka.md` > "Better to ask forgiveness than permission."\ > — `2026-02-13 Shantanu Bhanja Ashoka.md` **Interpretation** This captures a real tension. AI innovators want to move by experiment; universities often need a trail of **legitimizing artifacts** before a change becomes part of the institution. ### 12. Student readiness, cognitive load, and UI clutter can defeat even well-meant AI interventions The OPPE discussions show that some problems are neither Senate nor IT. They are **design and support failures at scale**. **How it gets stopped** - students hit cognitive overload - UI clutter may itself be creating failure - staff struggle to tell whether the failure is ability, design, or stress - scale makes diagnosis hard **Quotes** > "What is the impact of the entire UI/UX? ... we are not sure whether there is some kind of cognitive load because of the interface..."\ > — `2026-03-11 IITM BS Python OPPE.md` > "Graded programming assignments is mandatory because it affects their eligibility. So if they don't do graded programming assignments, they are not allowed to sit in OPPEs these days."\ > — `2026-03-11 IITM BS Python OPPE.md` > "Around 3900 students... around 4000 of them have done a non-proctored programming exam."\ > — `2026-03-11 IITM BS Python OPPE.md` **Interpretation** This matters to governance because poor UX and high cognitive load later show up as complaints, failures, and fairness disputes — which committees then have to manage. ### 13. Procurement and RFP processes are badly matched to fast-moving AI systems The SWAYAM / IITM RFP transcript makes this explicit. **How it gets stopped** - long-term RFPs assume stable requirements - model changes and workflow changes are constant - explainability and editability are not optional, but procurement tends to under-specify them **Quotes** > "SWAYAM will need to explain to parents why AI is grading their children."\ > — `2025-12-18 AI for Swayam IITM RFP.md` > "Requirements WILL change. AI evolves. Adoption evolves."\ > — `2025-12-18 AI for Swayam IITM RFP.md` > "Long term RFPs are singularly unsuited for AI projects given the fast pace of change."\ > — `2025-12-18 AI for Swayam IITM RFP.md` **Interpretation** This is a very practical governance pain point: **the institution's buying process is slower and more static than the technology it is trying to buy**. ### 14. Underneath all of this is employability panic: institutions know graduate relevance is at risk This is the deeper reason these conversations feel urgent. **How it gets stopped** - institutions know jobs are changing - but curriculum change is slower than labor-market change - this creates pressure, anxiety, and sometimes overreaction **Quotes** > "What happens to engineering institutions? This is not an NIE-only problem... this is a global problem..."\ > — `2026-03-12-nie-ai-roadmap/transcript.md` > "Basically, a lot of the engineering sciences, STEM sciences, are under threat of big change."\ > — `2026-03-12-nie-ai-roadmap/transcript.md` > "AI is going to put a lot of people out of relevance."\ > — `2026-02-07 Vishnu.md` **Interpretation** This is not itself a Senate blocker, but it is the background pressure that makes governance difficult: institutions are being asked to redesign for a future that feels urgent but still politically uncertain. ### 15. Some faculty fear AI may arrive too early, weaken fundamentals, or harm student well-being Not every objection is conservative obstruction. Some are genuine concerns about **sequence, cognition, and dependency**. **How it gets stopped** - faculty worry students will use AI before they build fundamentals - AI is compared to giving a calculator before children learn tables - there is concern about student and faculty well-being under constant technology use **Quotes** > "Second question is about introducing AI in first year... it's like asking a child to learn tables giving them a calculator. Giving the calculator first. So naturally people will take the easy path."\ > — `2026-03-12-nie-ai-roadmap/transcript.md` > "How about student's well-being and cognitive knowledge in use of this technology in day-to-day life? ... parallelly we need to see the well-being of students also. Well-being of students as well as faculty members."\ > — `2026-03-12-nie-ai-roadmap/transcript.md` > "It is not enough if we say, 'Oh there is wonderful content, there is AI, ask ChatGPT, you learn yourself.' I don't think it works."\ > — `2026-03-12-nie-ai-roadmap/transcript.md` **Interpretation** This matters because some institutional hesitation is not anti-AI. It is a demand for a **sequenced pedagogy**: where AI enters, what fundamentals remain non-negotiable, and how student well-being is protected. ## Bottom line Across these transcripts, the biggest lesson is: **The hard part of AI innovation in education is rarely the model. It is the institution.** People are blocked by: - skeptical management committees - Senate statutes and approval fatigue - platform and recording rules - IT/security restrictions - lack of TA/support capacity - integrity arms races - broken source-of-truth data - faculty incentive problems - procurement processes built for slower technologies So if an Academic Council wants to move seriously on AI, it cannot treat this as only a teaching-method problem. It has to treat it as a combined problem of: - curriculum - assessment - governance - infrastructure - support - and institutional change management