How an Instructional Designer With No Technical Background Built a Working AI Agent Through the Johns Hopkins Agentic AI Program

An Instructional Designer Who Decided to Understand AI From the Inside

ES

Elba M. Sepulveda

Instructional Designer

Elba has spent her career making sure learning actually works. She sits in the room where a course is built from scratch, where faculty try to figure out what students should walk away knowing. When AI started showing up in those conversations and nobody had good answers, she stopped waiting for someone else to figure it out.

She Was Tired of Being in the Room Without the Right Tools

It started in meetings. The same questions kept coming up and nobody had satisfying answers. How do you bring AI tools into course design without gutting what students are actually supposed to learn? How do you make thoughtful decisions about technology that is moving faster than most institutions can track? Elba kept showing up to those conversations and realised the same thing every time: knowing the buzzwords was not going to cut it. She needed to get inside the work.

She enrolled in the Certificate Program in Agentic AI by Johns Hopkins University with Great Learning because she had a real problem to solve and needed real skills to solve it.

“

I wanted to explore the AI-driven approaches to enrich instruction and provide them with some solutions.

E

Elba M. Sepulveda On her motivation for enrolling

The program took her through LangChain, Python notebooks, JSON, agent architecture, and tool integration. Each week built on the one before, and the material never drifted into abstraction. Everything pointed back to something she could actually use.

She Saw a Slow, Broken Process and Built Something to Fix It

The Course Design Matrix is one of the most important documents in higher education and one of the most frustrating to produce. It is where learning objectives get mapped to activities, assessments get connected to outcomes, and the entire logic of a course gets worked out before a single student ever sees it. It is also slow, inconsistent, and completely dependent on how much experience the instructor sitting in front of it happens to have that day.

Elba saw that as a problem worth solving. The agent she built does not do the work for the instructor. That was a deliberate choice.

“

The goal of the agent is not to create the whole course or to fill out the document for them, as it can do now, but to provide the conversation and guiding questions to facilitate the process.

E

Elba M. Sepulveda On the design intent behind her agent

The agent asks questions, listens, and responds to what it hears. It generates learning objectives from a course description, aligns them across modules, and suggests activities that are shaped by what the instructor has actually said, not by some generic template. The instructor stays in the driving seat. The agent just makes the journey faster and better structured.

“

These types of activities are essential for real-world applications and cannot be achieved solely through GenAI, as they require authentic human context and interaction.

E

Elba M. Sepulveda On the boundaries of AI in education

The Part Nobody Tells You: You Learn by Building, Not by Watching

The mentor sessions were where she could get stuck on real problems and work through them with someone who knew the material. Masterclasses connected what she was building in code to how these tools actually get used in the field. The projects at the end of each section were not extras. They were the point, the moment where everything she had been learning had to actually work.

When her code broke, she used the same AI tools she was learning to build. She would describe the error, follow the explanation, and come out of it understanding not just the fix but why the fix was right. Learning and building were happening at the same time, each one feeding the other.

“

The GenAI tool served as a personal tutor, allowing you to troubleshoot and helping you better understand not only the process but also the code itself.

E

Elba M. Sepulveda On using AI tools to learn AI

She finished the program with real technical skills and a clear sense of what she wants to build next.

“

I also see the need for a second phase. One that explores broader deployment opportunities and introduces simpler, no-code solutions for building agents and deploying them into other scenarios.

E

Elba M. Sepulveda On what she wants to build next

Eight Hours a Week, a Family Crisis Midway Through, and She Still Finished

Eight hours a week, minimum. Lectures during lunch. Coursework after the kids were in bed. Quizzes on Saturday mornings. Notebooks on Sunday afternoons. Elba built a routine and held to it, even when life made that much harder than she expected.

Halfway through the program, her sister needed her. Elba was travelling regularly to help out. The online format meant she could keep going, fitting study into whatever gaps she could find, doing the work in hotel rooms and on borrowed time.

“

I need to be honest: it was very challenging at the beginning because I was adding new responsibilities to my already full schedule.

E

Elba M. Sepulveda On the honest difficulty of the commitment

“

This experience taught me resilience and reinforced the importance of structure when balancing professional development with personal and work responsibilities.

E

Elba M. Sepulveda On what the program taught her beyond the technical

Her Advice Is Honest in a Way Most People Are Not

Elba does not tell everyone to build an agent. She is specific about when it is worth the effort and when it is not.

“

If the task does not require complex decision-making, a RAG process with an LLM can often handle it effectively. If you don’t have a clear goal for the agent, it’s perfectly fine not to build one, as unnecessary development can waste resources.

E

Elba M. Sepulveda On when to build an agent and when not to

What she is unambiguous about is this: you have to do the work with your hands. Reading about these tools and building with them are two completely different experiences, and the distance between them is larger than most people expect before they start.

“

Engage in hands-on learning experiences like this one to truly understand the process. Working with others can significantly reduce effort and lead to results that exceed expectations.

E

Elba M. Sepulveda Her advice to professionals on the fence

What the Program Covers

The curriculum runs from the foundations through to building and deploying a working agent, with every module tied directly to something you can apply. Nothing is left hanging in theory.

• 01
  Agent Architecture & Foundations
• 02
  LangChain & Tool Integration
• 03
  Python Notebooks & JSON for AI Development
• 04
  Designing AI Agents & Workflows
• 05
  Tool Integration & Workflow Orchestration
• 06
  Debugging & Iteration Using GenAI Tools
• 07
  RAG Processes & When to Use Them vs Agents
• 08
  Capstone Project: Building a Functional AI Agent

Technical and Professional Capabilities Gained

Participants come out with real technical skills and the professional judgment to know where and how to use them. The two tend to develop together.

Strengths and Considerations

Who This Program Is Designed For

If you are tired of watching AI change your field from the sidelines, this program was built for you. Here is who gets the most from it:

• Instructional designers looking to integrate AI meaningfully into course development
• Education professionals working on curriculum, learning experience design, or faculty support
• Non-technical professionals who want hands-on AI skills without a developer’s background
• Professionals in any field where AI is reshaping workflows faster than organisations can respond
• Learners who want to build real solutions, not just understand concepts
• Working professionals who need a flexible structure that fits around a full schedule

If you want an overview rather than the real work, this is probably not the right fit. The program asks something of you.

Common Questions About the Program