In Spring 2025, I redesigned Introduction to Entrepreneurship at the UC Merced School of Engineering around the use of AI tools for entrepreneurs. We weren’t building models from scratch. We were doing what we as a society are doing—trying to figure out the more productive application of AI, in this case for startups specifically.
A course “about AI” can remain entirely theoretical. A conventional entrepreneurship course with an added prompt-writing exercise can leave the underlying work unchanged. I wanted to examine what happens when AI is available throughout the entrepreneurial process—and when students are required to judge its output rather than merely produce it.
AI literacy is workflow literacy
Knowing how to ask a model a question is not the same as knowing how to use AI well. Useful work requires framing a task, supplying context, choosing a tool, decomposing a problem, inspecting intermediate output, verifying claims, and deciding what should remain a human responsibility.
Entrepreneurship makes those requirements unusually visible. The work crosses domains: customer research, market structure, pricing, positioning, operations, financial reasoning, communication, and the construction of agreements with other people. A single confident answer cannot substitute for understanding how these parts fit together.
The educational objective therefore cannot be “learn the right prompts.” Prompting techniques change quickly and are easy to imitate. The durable skill is building and evaluating a workflow whose output can support action.
Fluency changes the nature of novice work
Generative systems can give a novice an immediate first draft of almost anything: a market analysis, interview guide, value proposition, launch plan, financial explanation, slide outline, or email. This is genuinely useful. The blank page becomes less intimidating, unfamiliar vocabulary becomes more accessible, and students can compare several alternatives quickly.
But fluent first drafts create a new problem. Weak work no longer announces itself through incomplete sentences or an empty page. It can arrive polished, organized, and wrong.
That shifts part of teaching from production to discrimination. Students need to ask: What evidence supports this? Which assumptions were supplied by the model? What information is missing? Does the recommendation follow from the facts? Is this merely a generic pattern that could describe almost any venture? Who is responsible if the answer is used?
Lowering the cost of articulation raises the importance of judgment.
Entrepreneurship is not generated in a chat window
AI can help analyze a market, produce candidate language, identify comparable businesses, summarize regulations, or suggest experiments. It cannot by itself enroll a collaborator, establish trust with a customer, negotiate access to a resource, or create the rights and obligations that make an organization possible.
This is especially important to my approach to entrepreneurship. New economic possibilities are often constructed through agreements: a supplier grants access, a partner contributes a capability, a customer commits to participate, or an institution recognizes a new role. These are changes in the social world, not merely descriptions of possible changes.
AI can help an entrepreneur think through and articulate an agreement. The agreement becomes real only when people recognize it and accept the associated rights, duties, permissions, authority, or commitments.
That is a useful limit for students to encounter. AI can expand the space of candidate action. Entrepreneurship still requires action in a world shared with other people.
Evaluation must be built into the assignment
If AI use is permitted but evaluation is optional, the easiest path is to submit the first plausible output. A serious AI-intensive course has to make evaluation part of the work itself.
The student should be able to explain what a tool contributed, what was checked independently, what changed after testing, and why the final decision is defensible. Unsupported output should not become acceptable merely because a model produced it. Nor should students be rewarded for hiding AI use when transparent documentation would make the reasoning easier to assess.
Banning a list of tools does not solve this. Assignment design and epistemic responsibility do.
The instructor’s role grows
When models can explain basic concepts instantly, instructors may be tempted to think that less teaching is required. I reached the opposite conclusion.
Students have more candidate explanations, more apparent evidence, and more polished alternatives than before. They need frameworks for deciding among them. They need examples of genuine expertise, causal explanation, and domain-specific judgment. They need to see why an answer that sounds sophisticated may still be irrelevant to the decision at hand.
The instructor is no longer the only available source of articulation. The instructor remains responsible for helping students distinguish articulation from understanding and output from justified action.
A course for an unstable technological moment
Any course organized around named AI products will age quickly. The models, interfaces, prices, and capabilities will change. A durable course should therefore teach students to evaluate classes of tools and kinds of tasks.
The central questions remain stable:
- What is the actual task?
- What information and judgment does it require?
- Which part can be delegated safely?
- How will the output be checked?
- What important context is absent from the model?
- What must happen in the world before the proposed venture becomes real?
Teaching entrepreneurship around AI did not make entrepreneurship a branch of software instruction. It made the structure of entrepreneurial judgment clearer. Students could generate more possibilities and articulate them faster. The educational work shifted toward evaluation, selection, testing, coordination, and responsible commitment.
That is likely to be the broader pattern for AI in professional education: less time spent producing routine first drafts, and more attention to the human work required to determine which drafts deserve to shape reality.