From Vibe Coding to Guide Coding

AI-assisted development can give people with little technical experience a new way to bring ideas to life. What do students still need to learn when they can produce a polished app before understanding its purpose or possible consequences?

From the thrill of making to meaningful creative work

The chapter takes the initial excitement of making an app seriously. Lower barriers can open opportunities for creative expression. Yet an app can function before its builder has established whether it addresses a meaningful problem. Its effects on users may also remain unclear. Creative work requires originality and meaningfulness or acceptability within a particular context.

A polished interface alone cannot establish these qualities. Speed may also narrow the space for developing more original possibilities. The educational challenge is to help students understand what they are making and evaluate what it contributes. Guide coding is proposed as a way to keep students involved in the decisions that shape what their ideas become.

Introduction, PDF pp. 1–2.

Guide coding adds creative agentic friction at decisions that need human judgment. These pauses invite students to develop distinct possibilities and act on their ideas while retaining ownership of what those ideas become.

What makes the friction creative and agentic?

Guide coding is a form of teaching for students’ creative agency during AI-assisted development. Friction is creative when it invites students to frame a problem in a distinctive way or move beyond the first plausible output. It is agentic when students use that space to act on their ideas and remain responsible for what those ideas become.

Educators introduce these opportunities at consequential decision points. Dialogue with AI can help students clarify their purpose and examine alternatives. The resulting guidance supports both the student builder and the AI system by making the learner’s intentions explicit.

Guide Coding as a Special Form of Creative Teaching, PDF pp. 3–4.

Delegation can support creative agency

Slow AI begins with human purpose and judgment, with learners retaining responsibility for interpreting and using what AI produces. The duration of an exchange does not determine whether it follows this orientation. A brief interaction can draw on careful prior thinking. A long dialogue can still leave AI framing the problem and making the decisions.

The chapter distinguishes prompting for intellectual outputs from vibe coding for software artifacts. Agentic coding extends delegation to broader development workflows. These overlapping practices represent neither a required learning sequence nor an inevitable loss of agency. Strategic delegation may enable students to realize ideas they could not otherwise build.

Guide coding can support their choices across all these practices.

Fast and slow AI, PDF p. 3 and p. 7; trajectories of delegation, Table 1 and Figure 1, PDF pp. 4–6.

Prompting is an early place to learn how to delegate without surrendering judgment. Learners clarify what they want to explore, then question and verify the response before using it.

Purpose and dialogue before delegation

A simple question-and-answer interface can encourage learners to ask a question and accept whatever AI returns. In its most problematic form, this can become cognitive surrender, accepting an AI response without critical assessment. Guided prompting begins by clarifying the purpose of an interaction and providing relevant context.

Students can ask AI to challenge their thinking or offer another perspective. They also decide what to delegate and what remains their responsibility. Before accepting a response, they critique it and verify its claims. Before acting on it, they consider whether it reflects their judgment and what risks need attention.

Neither a direct exchange nor an extended dialogue determines educational value on its own. AI can provide a simulated perspective that learners question and move beyond while developing their own ideas.

Prompting and Guide-Coding Principles Applied to Prompting, PDF pp. 6–8.

Guide coding asks learners to explore a worthwhile problem and consider different ways of addressing it before committing to an app. Relevant knowledge and feedback can then inform a design specification, a written plan that guides development.

Keep the possibility space open

The first plausible prototype can narrow the possibilities a learner considers. The chapter draws on research about later ideas in creative thinking tasks and fixation on early designs to explain this concern. Students therefore begin with problem finding and relevant domain knowledge. For a creativity-learning app, they might investigate misconceptions about creativity and consider who needs help understanding them.

They generate different app concepts before choosing one to develop. Technical knowledge still matters because learners need to evaluate what AI produces and recognize when to intervene. A study discussed in the chapter illustrates how more experienced students supplied richer context to an AI coding system. It supports the need for informed engagement, without establishing that guide coding itself has been validated.

Vibe Coding and its guide-coding application, PDF pp. 8–10.

Build to learn and learn to create

Learners ground a proposed tool in relevant knowledge and seek feedback on the concept. They then develop a design specification, a written plan that states its purpose and intended users. The plan also describes the interaction flow and success criteria. Constraints and safeguards make the boundaries of development explicit.

During development, students compare the emerging tool with those intentions and discuss possible changes with AI. Beta-testing includes peers or likely users, alongside edge cases and failure scenarios. Deployment decisions address who can use the tool and what protections are needed. They also establish when to revise or remove it.

This cycle is recursive, so learners can revisit earlier decisions as their understanding develops.

Guide-coding Principles Applied to Vibe Coding, PDF pp. 9–11.

Agentic systems can plan and carry out sequences of development tasks on a learner’s behalf. Guide coding calls for checkpoints to examine this work, with people retaining responsibility for deciding whether the resulting tool should be used.

Guide the work before, during, and after delegation

As AI takes on broader workflows, students’ decisions can become less visible. The chapter raises a related paradox. More automation may increase the need for technical knowledge and careful human oversight. Before work begins, learners provide their purpose and domain grounding, then ask the system for a plan.

They also establish success criteria and non-negotiable safeguards. During development, checkpoints allow students to review proposed actions and request explanations of completed work. Co-creative dialogue gives them opportunities to question alternatives and redirect the system. After development, students remain responsible for further testing and deployment decisions.

Monitoring continues after release, with redesign or removal when unintended consequences emerge. Reusable instructions can carry guidance into later agentic workflows, but students still need to judge whether those workflows support their agency.

Agentic Coding and Beyond and its guide-coding application, PDF pp. 11–12.