AI Intrapreneurship: Educators as Frontline Innovators in the Age of Generative AI

Introduction

The numbers around AI usage in higher education tell a story of disruption outpacing response. A 2025 survey of over 1,000 UK undergraduates found that 92% of students now use AI tools in some form, up from 66% just one year prior, and 88% use generative AI specifically for assessments (Freeman, 2025). The Digital Education Council’s global survey of nearly 3,900 students across 16 countries reported 86% student usage, with 54% using AI weekly and nearly one in four daily (Digital Education Council, 2024). On the other side of the classroom, a national survey of US instructors found that 72% have used generative AI for at least one instructional purpose, yet only 14% feel confident in their ability to use it for teaching (Ruediger et al., 2024). Faculty adoption, however, has surged from 24% in 2023 to 49% in 2025 (Cengage Group, 2024, 2025). I believe the question is no longer whether AI has arrived in higher education; it is whether the people closest to teaching and learning have the support, tools, and institutional conditions to respond effectively. For many faculty, the daily reality is a set of urgent pedagogical problems — students submitting AI-generated work, assessments that no longer measure what they were designed to measure, a firehose of new tools with no guidance on which to adopt — and an institutional response that oftentimes has not caught up.

Institutions have tended to respond to generative AI with a piecemeal approach of policies, task forces, enterprise licenses, and detection software. The EDUCAUSE 2025 AI Landscape Study found that 57% of institutions now consider AI a strategic priority, yet only 22% of respondents had an institution-wide approach to AI strategy and only 39% of institutions had formal AI acceptable use policies (EDUCAUSE, 2025). Meanwhile, AI detection tools remain unreliable: a 2023 evaluation of 14 tools found accuracy ranging from 33% to 81% (Weber-Wulff et al., 2023), while in a subsequent study, false positive rates ranged from 0% to 32% across three detection tools (Pratama, 2025), and non-native English texts were flagged as AI-generated at higher rates than native English texts (Liang et al., 2023). All this contributes to a challenging environment for faculty and instructional designers seeking to ensure that students authentically achieve learning outcomes in courses. Asynchronous written assignments — essays, research papers, take-home exams, online discussion posts — can now be completed by generative AI tools that produce plausible work almost instantly (Kofinas et al., 2025). Even assessment designs once considered more robust, such as case study analyses that ask students to apply course concepts to realistic scenarios, can be completed by generative AI that synthesizes course readings and produces contextually appropriate responses. Agentic AI browsers and systems capable of navigating learning management systems and completing weeks of coursework with minimal student involvement may render even carefully designed assignments vulnerable, particularly in online modalities (Gulya, 2026). Personally, in the fall of 2025, I used OpenAI’s agentic Atlas browser to complete an entire assignment, titled “Entering the Conversation,” from my online advanced composition course. The tool navigated several different websites, selected an appropriate article from a list of over three hundred possibilities, copied three verbatim quotes from that article, and wrote a “personal response” to these quotes, discussing their potential application in a paper. The AI agent accurately reproduced all three quotes and created a discussion response that I believe was of a quality that could easily pass for student work without raising my suspicions, and all I had needed to do was paste in the assignment prompt and a few additional instructions; the AI did the rest.

The implications for student learning are stark. Early research on AI-permissive assessments, still in pre-print, found that students achieved near-perfect scores on modular assignments where AI use was permitted, while performance dropped by approximately 30 percentage points on proctored exams — a Cohen’s d of 1.51, indicating a significant AI inflation effect (Ding, 2026). While more research is needed to confirm these findings, for faculty and administrators who care about students genuinely achieving course and program learning outcomes, the challenge generative AI poses on its face to traditional modes of assessing learning is potentially disruptive and overwhelming. Institutions also face a bigger-picture challenge; if, in fact, students can simply complete most college work with no effort or genuine learning, the basic value proposition of higher education comes into question for students, who might realistically substitute inexpensive AI subscriptions instead, and for employers, who would then no longer comfortably assume a college degree is earned and signifies achievement or authentic expertise.

The scale of this disruption raises a fundamental question: how ought we as faculty and instructional designers respond? Standard technology adoption frameworks offer limited guidance. A grounded meta-analysis of 45 studies concluded that frameworks such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) “inadequately address the unique ethical, pedagogical, institutional, and technical complexities specific to AI” (Taheri et al., 2025). And the mismatch between institutional responses and faculty needs is structural, not incidental: Schophuizen and Kalz (2020) found that bottom-up educational innovation in universities generates local momentum but requires coordinated top-down support to scale. Yet top-down technology strategies risk what Lisewski (2004) describes as “organizational schizophrenia,” a mismatch between strategic goals and achievable practice, when they assume a unified institutional culture and overlook the fragmented, discipline-specific contexts in which faculty actually work. This structural tension suggests why institutional AI responses may not successfully address the specific challenges faculty face. In my view, this difficult situation argues for faculty and instructional designers to take a practical, proactive approach that draws on the existing concept of intrapreneurship.

I propose AI intrapreneurship as a framework for understanding what faculty and other educators are already doing, what they could be doing, and what institutional conditions enable or constrain them. Intrapreneurship — the act of behaving entrepreneurially within an existing organization (Pinchot, 1985) — is distinct from the top-down corporate entrepreneurship that Sharma and Chrisman (1999) distinguished it from: when a university launches an AI task force or purchases an enterprise license, that is corporate entrepreneurship; when a professor builds a custom chatbot for her students on a weekend, that is intrapreneurship. I define the concept as follows, with three specific modes of action:

AI intrapreneurship describes educators who take hands-on responsibility for (1) building AI-powered solutions for teaching and research, (2) building AI-resilient solutions that ensure authentic, verifiable student learning, and (3) reclaiming bandwidth by leveraging AI to reduce administrative burden.

All three of these modes share a behavioral core: in each, educators recognize that AI has changed their professional reality, take hands-on initiative to respond, and do so within, and sometimes despite, institutional contexts that have not kept pace. Indeed, the three modes work synergistically. Mode 3 (reclaiming bandwidth) creates the time for Modes 1 and 2. Mode 1 competency (building with AI) directly informs Mode 2 (building resilience to AI), because faculty who understand what AI can do are better positioned to design assessments it cannot shortcut. Once I saw what the Atlas agentic browser could do, for example, I had a better picture of the scope of the challenge of AI-resilient assignment design, if not any easy answers. Mode 2 then also generates pedagogical insight about AI capabilities, student behavior, and assessment design that feeds back into more effective Mode 1 innovation. To best understand these cycles, I think it is important to recognize the connection to previous digital transformations. Pinchot and Soltanifar (2021) extended intrapreneurship specifically to digital transformation contexts, coining “digital intrapreneurship” as the corporate solution to rapid digitalization; this article aims to bring that concept to bear on higher education’s teaching mission.

Yet intrapreneurship remains underexplored with respect to higher education faculty. A bibliometric analysis found that of 9,506 Scopus records for “intrapreneur,” only 712 intersect with entrepreneurial university literature, and virtually none address the teaching mission or AI-specific innovation (Gregán et al., 2024). The “entrepreneurial university” tradition (Clark, 1998; Etzkowitz, 2016) operates at the institutional level — Triple Helix partnerships, spin-off companies, industry engagement — and does not address individual faculty innovating within their classrooms. In what follows, I first make the theoretical case for AI intrapreneurship and then position educators as uniquely capable and newly enabled front-line innovators. The three central sections operationalize the three modes with concrete use cases. The Conclusion addresses tensions and institutional implications, including how faculty can use intrapreneurship theory to identify and advocate for the organizational conditions that support their work.

The Case for AI Intrapreneurship

Standard frameworks for understanding technology adoption in higher education describe the competencies faculty need and the barriers they face, but they cannot adequately explain why the vast majority of faculty who have experimented with AI still lack the confidence to integrate it meaningfully into their teaching (Ruediger et al., 2024). Taheri et al.'s (2025) grounded meta-analysis of 45 studies identified four barrier clusters — individual factors, infrastructure, tool factors, and impact concerns — and concluded that TAM- and UTAUT-based models fail to account for the unique complexities of AI in education. AI literacy frameworks tell a similar story of incompleteness. The EDUCAUSE AI Literacy in Teaching and Learning framework identifies competencies across technical understanding, evaluative skills, practical application, and ethical considerations (Kassorla et al., 2024); UNESCO’s AI Competency Framework for Teachers outlines 15 competencies across three progression levels (Miao & Cukurova, 2024); Allen and Kendeou’s (2024) ED-AI Lit framework adds collaboration, contextualization, and autonomy. These frameworks describe what educators should know. What they do not explain is how and why educators should move from knowledge to sustained action or what organizational conditions prevent that transition. The 72%–14% gap between experimentation and confidence that Ruediger et al. identified is not merely a skills deficit; it likely reflects, at least in part, organizational conditions that suppress the conversion of awareness into sustained, productive innovation.

Intrapreneurship theory maps precisely the terrain that adoption models and literacy frameworks leave uncharted: the organizational conditions under which individual initiative either translates into innovation or dies on the vine. The concept has a substantial track record in corporate settings. When 3M institutionalized its “15% time” policy, engineer Art Fry used that protected time to develop the Post-it Note — a product born from a problem no one in management had identified, built with a colleague’s “failed” adhesive, and navigated through years of institutional resistance before becoming one of 3M’s most iconic innovations. Google adapted the concept as “20% time,” which produced AdSense, now responsible for billions of dollars of annual revenue, and Google News, among other products (Page & Brin, 2004). More recently, Adobe’s Kickbox program gave every employee a physical innovation kit containing $1,000 in seed funding, a structured curriculum, and no approval requirement, deliberately removing the organizational barriers that stifle bottom-up innovation. The program has been adopted by over 1,000 organizations; at Siemens Energy alone, 220 employees received Kickbox kits, 57 projects advanced to the prototype stage, and 12 reached commercial viability (Innov8rs, 2021). What these examples share is not a specific technology or industry but a structural insight: when organizations create the conditions for bottom-up innovation — protected time, seed resources, tolerance for risk, and permeable boundaries — people closest to problems become the most effective solvers of those problems.

Neessen et al. (2019) formalized this insight into an integrated model identifying five behavioral dimensions of the intrapreneurial employee: innovativeness, proactiveness, risk-taking, opportunity recognition, and networking. These dimensions describe precisely what effective educator-innovators do. An instructor who recognizes that her online discussion board has been compromised by AI-generated posts is exercising opportunity recognition. When she acts before her department issues guidance, designing a replacement oral assessment format, that is proactiveness. Building and piloting that assessment despite uncertainty about whether it will succeed is risk-taking. Sharing her approach with colleagues in another department, creating a template others can adapt, constitutes networking. And the assessment design itself — something no existing vendor product addresses for her discipline — represents innovativeness. Each of these behaviors is individually necessary; together, they constitute the intrapreneurial disposition that distinguishes sustained innovation from casual experimentation.

The case for applying this framework specifically for educators rests on both empirical evidence and institutional logic. Faculty who adopt AI are motivated by efficiency (52%), becoming better instructors (44%), boosting creativity (42%), and being early adopters (42%) (Cengage Group, 2023), the last category mapping directly onto Rogers’ (2003) innovation diffusion categories and intrapreneurial proactiveness. Cabero-Almenara et al. (2024) found that constructivist pedagogical beliefs significantly predict AI adoption, while Chen et al. (2025) demonstrated that high AI self-efficacy can convert anxiety into productive motivation. Moraes et al. (2025) studied 680 professors across more than 70 Brazilian universities and found that university support strongly influences faculty intrapreneurship. Yet the intrapreneurship literature seems so far not to have meaningfully touched the teaching mission; I found no application to AI-specific faculty innovation, and the handful of existing studies focus on research commercialization or institutional strategy rather than classroom-level innovation (Burkholder & Hulsink, 2025; Nugent & Lambert, 1994; Smith et al., 2014). Non-tenure-track faculty, instructional designers, and community college instructors — who collectively serve the majority of students — are also largely absent from this literature.

The framework I advance here applies to educators who exercise pedagogical judgment and course-design authority. For the individual faculty member or instructional designer navigating AI’s disruption, intrapreneurship offers something that literacy frameworks and adoption models do not: a language for understanding their own innovative behavior as legitimate professional work rather than unauthorized tinkering, a basis for advocating for the institutional support they need, and a framework that connects their classroom-level efforts to a broader tradition of bottom-up innovation with demonstrated results across industries.

Mode 1: Educators as Front-Line Innovators

Faculty possess a unique combination of domain expertise, proximity to learners, pedagogical content knowledge, and course-design autonomy that no centralized office can replicate. But the argument for educator-led AI innovation does not rest on the claim that faculty knowledge is irreplaceable; frontier AI models are increasingly capable of sophisticated content generation and feedback. The stronger claim is that we faculty are uniquely positioned as problem identifiers and design decision-makers; we see the gaps between what students need and what existing tools provide, we understand the pedagogical context in which tools must operate, and we exercise the judgment required to decide what should be built, for whom, and why. This is the intrapreneur’s core competency — opportunity recognition in context. Educators who communicate for a living and routinely translate between expert and novice understanding are well-suited to the kind of natural-language interaction that AI-assisted building requires; the pedagogical skill of formulating clear, contextualized explanations translates directly into the structured prompting that vibe coding and no-code AI platforms reward.

What has changed in the past two years is that AI has not only created the need for faculty intrapreneurship but also dramatically lowered the bar for practicing it. The emergence of “vibe coding” — a term coined by Andrej Karpathy in February 2025 to describe building software through natural language prompts — represents an inflection point; the barrier to building functional software tools has dropped from years of programming training to the ability to describe what you need in plain language. Some institutions have sought to leverage this capability to empower faculty; for example, MIT Sloan’s Teaching & Learning Technologies team introduced Stack AI, a no-code platform enabling faculty to build course chatbots, interactive simulations, and AI-enhanced student projects without coding (Alvarez & Silvestrone, 2024). EDUCAUSE has offered learning labs specifically for designing custom AI assistants, requiring no prior coding experience and culminating in participants building working chatbot prototypes for their own institutional contexts (EDUCAUSE, 2024). Gartner predicts that non-professional builders using low-code and no-code platforms will outnumber professional developers 4:1 by 2026 (Muhammad et al., 2025). The World Economic Forum’s (2026) workforce report explicitly recommends that organizations “recognize intrapreneurship and innovation behaviours” in AI-driven workforce transformation, providing direct external validation for the framework this article advances. This development addresses a traditional limitation of the intrapreneurship framework and its applicability to higher education faculty who may not have disciplinary or practical experience in either entrepreneurship or intrapreneurship; intrapreneurship, particularly around technology, has historically required technical skill or organizational capital, which limited it to a small subset of employees. Now, AI-assisted building tools can largely dissolve that constraint for educators, enabling faculty who could never have written code to prototype functional tools in hours, and, in doing so, to solve pedagogical problems that were previously unsolvable without a development team or a vendor willing to build for a niche market. AI creates the disruption and at least the partial means to respond to it, a dynamic that traditional intrapreneurship theory did not anticipate but that the framework accommodates naturally.

Faculty who practice Mode 1 intrapreneurship can create AI-powered tools that solve problems in teaching and research, problems that are visible from the front line but often invisible to centralized IT and administration. For instance, some promising applications are those that are highly personalizable and can meet students where they are in ways that a single instructor cannot: course chatbots grounded in discipline-specific content that serve as 24/7 study partners, FAQ handlers, and formative assessment tools; interactive simulations built with domain knowledge that no generic product could replicate, like clinical decision-making scenarios in nursing education, historical role-plays in humanities courses, engineering design challenges; and content generation tools that produce practice problems, study guides, and multi-modal reformatted materials (text-to-audio, accessible format converters) at a scale a single faculty member could never achieve alone. MIT Sloan’s Stack AI deployment provides a reference: faculty-built course-specific chatbots and simulations without writing code, using no-code platforms to encode their disciplinary knowledge into tools that function as extensions of their teaching (Alvarez & Silvestrone, 2024). The intrapreneurial dimension is not the use of AI but the act of creation, recognizing a pedagogical gap, taking initiative to address it, and navigating institutional constraints to build and sometimes deploy a working solution. When an educator uses ChatGPT to draft a syllabus, they are using a tool; when they build a custom GPT trained on their course readings that generates Socratic questions calibrated to their students’ common misconceptions, they are practicing intrapreneurship.

An abiding concern from institutions tends to be security, and it is important to acknowledge the real risks of vibe coding and citizen development. For instance, Veracode’s 2025 analysis found that only 55% of AI-generated code is secure; still, the citizen developer literature prescribes governance frameworks and IT collaboration as the appropriate institutional response (Muhammad et al., 2025). Concerns about security, while often legitimate, can also stem from a bureaucracy’s resistance to change, which might limit needed innovation from intrapreneurs. Working together is important, however, because the “last mile” problem remains significant; a single faculty member or instructional designer building a prototype is not the same as deploying a tool at scale. LMS integration, FERPA compliance, accessibility standards, and long-term maintenance all require institutional partnership. Vibe coding lowers the entry barrier to prototyping; it does not eliminate the need for IT governance. Still, Adobe’s Kickbox program succeeded in part because it eliminated approval gates at the earliest stages of innovation; employees could spend the $1,000 however they chose, no questions asked. Universities that want to enable Mode 1 intrapreneurship need analogous mechanisms: lightweight processes for faculty to prototype and test AI tools, IT partnership models that treat faculty-built tools as innovations to support rather than shadow IT to suppress, and recognition that the educator who builds a course chatbot over a weekend is doing exactly what intrapreneurship theory describes and doing so in an institutional context that often may not recognize the effort. For faculty and instructional designers, intrapreneurship represents a chance to address the AI disruption in creative ways, rather than waiting for slow-moving institutions to solve the problems top-down.

Mode 2: Building Resilience to AI

I believe the assessment crisis created by generative AI cannot be solved by detection, policy, or any single institutional response, and the educators who have recognized this are already inventing solutions. Most if not all asynchronous written assessments are now highly vulnerable to shortcutting by general-purpose models. Detection tools have serious limitations, with accuracy ranging from 33% to 81% and false-positive rates spanning 0% to 32% depending on the tools used (Pratama, 2025; Weber-Wulff et al., 2023), with non-native English speakers disproportionately affected (Liang et al., 2023). The literature converges decisively on assessment redesign, not detection, as the sustainable path forward (Chan & Colloton, 2024; Perkins et al., 2024). This redesign work is a defining example of Mode 2 intrapreneurship.

The most robust AI-resilient strategies share a common principle: they require students to demonstrate understanding in ways that AI cannot simulate. Proctored in-person examinations, handwritten by students, verify human authorship by construction. The Two-Lane Approach, adopted by Australia’s TEQSA (Liu & Bridgeman, 2023; Lodge et al., 2023), provides a framework that acknowledges both AI-free verification and AI-engaged professional development: Lane 1, secure assessments like in-person proctored exams, assures foundational competency in supervised environments, while Lane 2, open assessments like projects or portfolios that may incorporate the scaffolded use of AI, develops professional capability where AI engagement is transparent and the human contribution is measurable. As an example of Lane 1, oral examinations are naturally AI-resistant because students must think in real time and respond to follow-up probes that test understanding rather than recall (Hartmann, 2025). These lanes may also be combined, of course; a student who submits a written analysis might be asked in a 15-minute oral defense to explain their reasoning process, respond to counterarguments, or extend their analysis to a new scenario. The intrapreneurial educator will leverage their own creativity, disciplinary knowledge, and innovative approach to find useful assessments and assignments that drive genuine learning for students.

Indeed, faculty intrapreneurs continue to experiment with new and modified approaches to drive learning. Process-based designs, such as portfolios with documented draft histories, AI-use logs, and metacognitive reflections, make the learning trajectory itself the assessable artifact. And AI-inclusive assessments, where students must demonstrate critical evaluation of AI outputs (for instance, prompting an AI to generate a literary analysis, then writing a critique identifying what the AI missed, oversimplified, or got wrong), might help to make the human contribution identifiable and measurable. The continued desire for transparency around expectations for AI use has also motivated the development of solutions like the AI Assessment Scale (Perkins et al., 2024), which provides a five-level framework for calibrating AI use across assessment types, running from no AI use in assessments to “full AI use.” Meanwhile, faculty training workshops that include hands-on AI use can tend to produce more effective AI-resilient designs (Awadallah Alkouk & Khlaif, 2024), suggesting that our intrapreneurship Mode 1, gaining competency through AI use, can indeed power effectiveness in Mode 2, designing AI-resilient assignments. This redesign work requires exactly the qualities Neessen et al. (2019) identify as intrapreneurial: creativity in inventing new assessment forms, proactiveness in acting before institutional mandates, and risk-taking in departing from established practices under uncertain outcomes.

Mode 3: Reclaiming Bandwidth through AI

The most persistent barrier to faculty and innovation is likely not skill, in my view, but time. Higher education faculty are expected to teach, research, serve on committees, advise students, and manage administrative compliance simultaneously, and the administrative share of that workload has grown steadily. Time availability is one of Hornsby et al.'s (2002) five organizational enablers of intrapreneurship, and it is the enabler most obviously in deficit in academic settings. Mode 3 addresses this directly: faculty who leverage AI to reduce the time consumed by administrative tasks reclaim bandwidth for the sustained, deep engagement that Modes 1 and 2 require. The gap between 72% experimentation and 14% confidence may partly reflect this: faculty who have tried AI tools and found them promising may simply lack the uninterrupted time to develop genuine competency. Google’s 20% time worked because it was protected; without that protection, it is easy to imagine the innovation trajectory of that company going differently. The same dynamic operates for faculty; without protected time, the intrapreneurial impulse risks getting snuffed out early. Support in terms of the supply of personal bandwidth could be meaningful for intrapreneurs to help drive impact, though perhaps not forthcoming from institutions.

This makes the advent of AI’s increasingly productive capabilities especially valuable for educators to understand and to leverage to claim back bandwidth. Agentic AI — systems that pursue multi-step goals autonomously, using planning, memory, and tool integration (Acharya et al., 2025) — represents a qualitative leap beyond chatbot interactions and offers the most promising pathway for administrative bandwidth reclamation. Practical applications take many forms, though it is important to acknowledge that not all are equally accessible to every faculty member given current technology and institutional data infrastructure. Nevertheless, agentic AI and vibe coding provide a means of innovation limited mostly by the intrapreneur’s imagination. Examples of time-saving potential use cases abound: AI-assisted responses to routine student inquiries like syllabus questions, deadline confirmations, and policy clarifications, with escalation to instructors for complex cases, could represent relatively straightforward applications. With respect to the administrative burden faculty face, using AI tools to automate compliance and accreditation reporting, for example, could reduce the burden on program and department leads (Schroeder, 2026). Other potentially automatable processes like completion grading, rubric-aligned preliminary sorting, and scheduling automation could help to address tasks that consume faculty time without requiring pedagogical judgment. Less obvious sources of time drain can also be addressed with AI; for example, using Claude Code, an AI tool from Anthropic, I vibe-coded a link-checking Python application that automatically checked each of 321 links on the resource page of articles used by my composition students as the starting point for research papers. I should note that I do not know how to code, but now this task that might take an hour or two every six months can now be handled automatically. While this savings is not earth-shattering on its own, educator intrapreneurs likely could identify a wide range of small-bore opportunities like that, as well as others at a larger scale, that, taken collectively, could represent meaningful time savings without the need for administrative action to protect faculty time.

That said, some institutional implementations to streamline administrative issues are emerging; for example, Ithaca College’s Aurora system guides students through administrative processes (EAB, 2025), and Georgia Southern’s GUS provides centralized student communication (Kelly, 2025). These are student-facing tools that leave faculty administrative burden largely untouched, of course, illustrating how even institutions that invest in AI may direct those investments toward enrollment and student services and less toward the faculty bandwidth that drives pedagogical innovation. The intrapreneurial version of Mode 3 is the faculty member who, without waiting for institutional IT to act, builds or configures an AI workflow that automates their own administrative friction, a move that requires the same opportunity recognition and proactiveness that characterize Modes 1 and 2, if on a smaller and more personal scale. This reframing has strategic implications for institutions. Universities that deploy AI exclusively for student-facing services — recruitment chatbots, advising agents, application processing — might miss an opportunity to invest in the innovation capacity of their faculty. When an instructor spends three fewer hours per week on, for example, compliance paperwork, those hours become available for assessment redesign, tool building, and pedagogical experimentation. The return is not just administrative efficiency but the innovation capacity of the institution. The risks deserve honest acknowledgment: while Mode 3 focuses on automating administrative tasks rather than pedagogical interactions, faculty who adopt agentic AI tools should be attentive to the boundary between administrative and pedagogical work. Research suggests that when AI automation extends into learning-related interactions without adequate metacognitive scaffolding, students’ self-regulated learning capacity can be diminished (Xu et al., 2025), underscoring the importance of faculty judgment in deciding which tasks to automate and which to preserve as human interactions. Governance requirements around transparency, human-in-the-loop protocols, and data privacy are also substantial. Mode 3 is the most institutionally dependent of the three modes; many bandwidth reclamation applications require data access and system integration that individual faculty cannot achieve alone. The educator-intrapreneur must exercise judgment about where automation serves learning and where it undermines it, but the institution must also recognize that freeing faculty time is an investment in innovation, not merely an operational efficiency.

Conclusion

The three modes of AI intrapreneurship I advance in this article — building with AI for teaching, building resilience to AI through assessment redesign, and reclaiming bandwidth by offloading administrative tasks to AI — are connected responses to a single disruption. They are united by a common behavioral core: in each mode, educators recognize that generative AI has changed their professional reality, take hands-on initiative to respond, and navigate institutional contexts designed for a pre-AI world. What makes this moment distinct from previous waves of educational technology adoption is that AI simultaneously creates the urgency for intrapreneurship and lowers the bar for practicing it; vibe coding and no-code platforms enable faculty to build solutions that would have required a software development team two years ago. The intrapreneurial educator is not a hypothetical figure; they are the professor who builds a course chatbot, the instructor who redesigns a compromised assessment, and the department chair who automates accreditation reporting so faculty can spend that time on teaching.

Still, a fundamental tension of the intrapreneurial approach, particularly in the academic context, lies between autonomy and coordination; faculty autonomy enables experimentation but also potentially creates siloed, invisible innovation. Schophuizen and Kalz (2020) found that bottom-up initiatives create awareness and find local resources but require synchronized top-down action to scale. Singh & Strzelecki (2026) found that observability is non-significant in faculty AI adoption; often, faculty are innovating in isolation, which means their innovations cannot spread. Communities of practice (Wenger, 1998) offer a promising mechanism for making invisible innovation visible and transferable, particularly when they are self-governed but strategically supported with targeted feedback and student success data (Hoyert & O’Dell, 2019), though sustaining such communities in academic settings remains a persistent challenge. The tension between AI-resilience and AI-integration requires navigating rather than resolving; the Two-Lane model acknowledges that both are necessary, and the intrapreneurial educator exercises judgment across that spectrum. Meanwhile, the standoff between individual agency and institutional inertia may be the hardest to address; promotion and tenure systems may not reward pedagogical innovation, given an institutional focus on research (Murray et al., 2019).

Faculty intrapreneurs who find their innovations stalling can use Hornsby et al.'s (2002) Corporate Entrepreneurship Assessment Instrument (CEAI) to identify the specific organizational barriers they face, and to articulate what they need, in language administrators recognize. The CEAI identifies five organizational enablers of intrapreneurship, and this article has mapped specific deficits to specific modes. Mode 1 requires work discretion and flexible organizational boundaries: IT partnership models that support faculty-built tools, procurement processes that accommodate prototypes, and permeable boundaries between faculty and instructional design units. Mode 2 requires rewards and reinforcement: explicit recognition for assessment redesign as scholarly work, so that faculty who innovate are not penalized in promotion or tenure for time not spent on traditional research. Mode 3 requires institutional investment in time availability: not merely encouraging faculty to experiment but actively reducing the administrative burden that prevents sustained engagement. An intrapreneurship-positive university culture would look something like Adobe’s Kickbox writ academic: seed funding for pedagogical innovation, structured but non-gatekept pathways for prototyping, protected time for experimentation, and visible celebration of faculty who build. Future research should empirically test the three-mode construct across institution types and appointment categories, examine what sustains intrapreneurial behavior over time, and compare AI-resilient assessment strategies across disciplines. Adaptation of the CEAI instrument for academic AI innovation contexts could provide institutions with a validated diagnostic tool. The educators who are already building AI-powered tools, redesigning compromised assessments, and automating administrative work to reclaim their time are not waiting for permission. They are recognizing opportunities, taking risks, and navigating institutional constraints to create innovations that serve their students. Institutions that wish to thrive in the age of generative AI must learn to recognize, support, and scale this work — or risk being innovated around by their own faculty.