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 value within a particular context.

A polished interface alone cannot establish either. 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.

Creative agentic friction involves deliberate pauses that give learners opportunities to shape AI-assisted work. Guide coding places these pauses at decisions that need human judgment, with the guidance changing as learners delegate more to AI.

What does it mean to slow down?

Creative agentic friction means pausing to bring human purpose and developing knowledge into decisions that automation could otherwise bypass. It draws on a Slow AI orientation, which begins with human thinking and returns responsibility for interpreting and using AI output to the learner. The length of an exchange does not tell us how much thinking it involves.

A brief exchange can follow careful preparation, while a long exchange can still leave AI shaping how a problem is understood. Guide coding applies this orientation through written plans and review. Learners may begin by requesting responses, then move to building tools. They may eventually delegate whole sequences of tasks.

At each point, guide coding calls for opportunities to examine decisions and retain responsibility for them.

Guide coding, PDF pp. 2–4; Table 1 and Figure 1, PDF pp. 5–6; discussion of fast and slow AI, PDF p. 8.

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 context before acceptance

A simple question-and-answer interface can encourage learners to ask a question and accept whatever AI returns. The chapter asks learners to consider what thinking they should do before consulting AI and what judgments they need to make afterward. They can provide relevant context and ask for an alternative perspective.

They can then examine whether the response extends or distorts their thinking, checking its accuracy before using it. A direct exchange can be appropriate. A longer dialogue does not, by itself, establish educational value. Guided prompting offers an early way to practice these habits before taking on the software-development work of guide coding.

The aim is to help learners collaborate with AI while continuing to shape the outcome with their own ideas.

Prompting and Guide-Coding Principles Applied to Prompting, PDF pp. 6–9.

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.

Build to learn, then examine what the tool contributes

The first plausible prototype can narrow the possibilities a learner considers. The chapter therefore suggests exploring the problem and possible approaches before starting to build. Learners draw on relevant knowledge to judge whether a proposed tool is worth developing. They seek feedback before turning the concept into a design specification that states its purpose and intended users.

This written plan also describes how the experience should work and what safeguards it needs. During development, students compare the emerging tool with those intentions. Testing includes likely users and situations where the tool may fail. The process is recursive, meaning that learners can return to earlier decisions as they learn more.

What they discover may lead them to revise the concept or reconsider whether the tool should be shared.

Vibe Coding and its guide-coding application, PDF pp. 9–12.

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 and remain responsible for its use

Before delegating work to AI, learners clarify their goals and the boundaries within which the system should work. During development, checkpoints give learners opportunities to inspect decisions and ask for explanations. Testing then informs whether to revise the tool or make it available for use. Responsibility continues through monitoring and, when needed, removing the tool.

The chapter suggests that greater automation can increase the need for informed oversight even as opportunities to exercise it become less obvious. Guide coding is proposed primarily for educational settings. Experienced developers may already use similar practices. Its value for learning and creative agency still needs to be studied.

Research should examine students’ understanding alongside the quality of what they produce.

Agentic Coding and Beyond and its guide-coding application, PDF pp. 12–14; Conclusion, PDF pp. 14–15.