Over the past two years, I’ve presented at multiple conferences (national and international), sharing the work I’ve been doing with AI-powered instructional design tools, multilingual learning platforms, and scenario-based tutoring systems. And something keeps happening when we mention AI in education: the room divides.

Some educators light up with excitement, while others lean back with visible discomfort, even distrust. I’ve had people tell me, “I don’t like AI,” “I don’t trust it,” or “I don’t think it belongs in education.” And I understand why they feel that way. If the only exposure someone has to AI is through an open chat interface like ChatGPT, Gemini, or Claude (used without guardrails, additional context, or instructional design structure), the results can be unpredictable. Large language models can be wrong, they can “hallucinate,” and they can generate answers that lack the nuance educators need. As teachers and researchers, we rely on accuracy and evidence, and those expectations don’t always align with the behavior of an off-the-shelf AI chatbot.

But here’s the part many educators don’t see: the possibilities that emerge when working behind the scenes through coding and APIs.

How We Shape AI Through Design

In my work over the last three years, everything we build (lesson planners, AI tutors, personalized exam practice activities, bilingual support tools) allows us to use AI in a controlled environment rather than a free-for-all conversation. Through APIs (the behind-the-scenes interfaces that allow software to communicate and follow precise instructions), we can give AI models strict instructions about what they can and cannot do, embed rubrics that define quality and expectations, supply rich pedagogical context drawn from subject-area standards and objectives, define users’ language proficiency levels in the target language, and suggest research-based instructional strategies appropriate for the specific content area and for emergent bilingual students. We can also verify reasoning with a second LLM, define explicit reasoning chains, embed subject-matter expert knowledge directly into the system, and even verify feedback using parallel analysis by multiple AI models before anything reaches a learner.

Researchers describe these methods as rich‑context prompting and multimodel verification loops that help ensure accuracy and reliability. When we use AI in this way (guided by design, supported by expert context, and reinforced through layered checks), it becomes far less like a “chatbot” and far more like a structured instructional engine that can be directed purposefully for educational goals.

Why Many Educators Distrust AI

The hesitation I see from professors, literacy specialists, and K–12 teachers is not a rejection of instructional innovation; it is a rejection of unpredictability. It is a reaction to the idea that AI makes decisions without pedagogical grounding. In my recent presentations, I show what becomes possible when AI systems are built with bilingual development rubrics, scenario-based learning patterns, alignment to learning objectives, equity-centered scaffolds, multilingual supports, controlled reasoning steps, and culturally responsive guidelines. I pair these with the technological layer that includes curated datasets, vector-based retrieval systems, embeddings, temperature control, multimodel verification loops, and other tools that shape how AI interprets and evaluates information. When educators see both sides working together, many begin to recognize that generative AI is not replacing pedagogy; it is enhancing it.

But here is the uncomfortable truth: we cannot build these systems correctly without educators and technologists working in true collaboration. And historically, education has struggled with this balance.

We’ve Seen This Before

When I worked as a classroom teacher, I saw how often technology departments made decisions without consulting teachers, bilingual specialists, or subject-area experts. Hardware and software tools were purchased without a clear understanding of how they aligned with curriculum, language needs, or the real conditions of classrooms.

Something similar happens in higher education, where education departments and technology departments (and their students) often move forward in isolation, working side-by-side (sometimes in the same building) without intersecting, collaborating, or meaningfully influencing one another’s work. The result is predictable: technologists build tools that don’t understand classrooms, and educators distrust tools that weren’t built with them in mind. Both sides want improvement, but they speak different professional languages.

AI Requires Collaboration

If there is one message I hope educators take from this moment, it’s this: Generative AI is too important to be left to technologists alone, and too powerful to be used by educators without the guidance of those who truly understand its inner workings and potential.

The future belongs to teams where teachers bring pedagogical wisdom, language experts provide insights about multilingual learners, literacy scholars contribute research-based frameworks, technologists offer architectural and coding expertise, instructional designers ensure structure and alignment, researchers bring evidence and evaluation, and AI specialists support model understanding.

For AI in education to be safe, effective, and equitable, it must emerge from genuine collaboration with educators rather than be imposed on them. It needs to reflect the realities of today’s diverse classrooms (spaces defined not only by multilingualism but by a wide range of cultural backgrounds, learning differences, and academic needs) and remain grounded in solid research on learning that accounts for this diversity across all subject areas. The systems we build must align with broad instructional goals and frameworks (such as national standards and state-level guidelines like CCSS, TEKS, ELPS, or WIDA) and be designed through the thoughtful combination of coding and instructional design that makes purposeful, effective use of APIs while incorporating safeguards, context, and structured reasoning. They also need to be tested with authentic student work and continually refined by subject‑matter experts who understand both pedagogy and the diverse needs of learners.

Signs of Success

The good news is that this collaboration is already happening in pockets, and the results are remarkable. I’ve seen districts and universities where educators and technologists co-design AI tools that teachers actually want to use. Tools that effectively support multilingual learners and enhance, rather than disrupt, teaching practices. But these success stories remain exceptions. We need this level of collaboration to become standard practice across all levels of education, from K-12 to higher education.

When I think back to the educators in my recent presentations who voiced their skepticism, I don’t interpret their hesitation as resistance or a lack of courage to try something new. It’s a reminder that the profession cares deeply about quality, equity, and responsibility. The goal is not to eliminate criticism but to invite collaboration. When educators and technologists work together, AI stops being a black box and becomes a tool shaped by human expertise, built for human learning and aligned with the needs of real classrooms.

I would love to hear your ideas about how we can begin building collaborations that bring educators, technologists, and researchers together to shape generative AI in meaningful and sustainable ways.