IND 598 · Arizona State University · Spring 2026
<aside>
FiveGuys
</aside>
<aside>
Every day, millions of people sit down to learn a complex digital tool, Figma, Unity, Adobe After Effects, Blender, and hit the same invisible wall. They know what they want to create. They do not know how to make the software do it. The default solution is to leave the tool, search YouTube, read a documentation page, return, try to remember what they just watched, and, more often than not, get lost again. This loop is not a skill problem. It is a design problem.
Homie AI addresses this problem directly. It is a real-time, in-context AI learning assistant that lives inside the user's workflow as a floating interface layer. Rather than sending users elsewhere for guidance, Homie delivers step-by-step visual instructions, contextual tips, and optional support modes, all without requiring the user to leave the tool they are trying to learn.
<aside> 💡
Value proposition: Homie AI transforms the learning experience from a cycle of context-switching and passive watching into active, guided practice inside the moment of work.
</aside>
The stakes are real. As design and development tools grow more powerful, they also grow more complex. Intermediate learners, the single largest group of tool users, are systematically underserved by both built-in help systems and external tutorial ecosystems. Solving this problem does not just help individual users: it accelerates onboarding for teams, reduces the skill gap in creative industries, and makes professional-grade tools genuinely accessible to people who cannot afford formal training.
Our evaluation of the prototype confirmed both the value and the direction. Users who tested the guided flow consistently described it as more actionable than any tutorial they had used before. The most striking finding was one we had not fully anticipated: completing a task with guidance does not automatically build confidence or retention. This insight reshaped our understanding of what a truly effective learning assistant must do, and it points directly toward the practice mode and scaffolded independence features that define Homie's next iteration.
<aside>
Fig-1. The Homie orb: the single persistent entry point that lives in the bottom-right corner of any complex tool.
</aside>
<aside>
Fig-2. Single tap reveals the four action buttons (screen sharing, add task, chat, voice); double-tap activates screen watching and the orb's eyes appear.
</aside>
Research in human-computer interaction has long established that learning is most effective when it occurs in context, a principle known as situated learning (Lave and Wenger, 1991). When instruction is embedded in the environment where the learner must act, cognitive load is reduced and transfer of knowledge is improved. This stands in direct contrast to the dominant model of tool learning today, which requires users to acquire knowledge in one context (a tutorial) and apply it in a completely different one (the software interface).
Cognitive load theory (Sweller, 1988) further explains why the current model fails. When a user must simultaneously hold tutorial instructions in working memory while navigating an unfamiliar interface, extraneous cognitive load consumes the mental resources that should be directed toward learning. In-context guidance reduces this load by presenting instruction and action in the same space. The concept of scaffolded learning (Vygotsky, 1978) also shapes Homie's design: the system is intended to offer structured support that gradually withdraws as the user's competence grows.
Several products attempt to address parts of this problem. Product walkthrough tools like Intercom and Appcues offer guided tooltips for onboarding flows, but they are built for SaaS products, not complex creative software. They also deliver static, pre-authored sequences that do not adapt to user behavior. General AI assistants such as ChatGPT can answer procedural questions about software, but they operate outside the tool and cannot see what the user is doing, forcing the user to translate text instructions into spatial interface actions, which is precisely where confusion originates.
Homie is distinct in three respects: it operates as an overlay layer that works across tools rather than being embedded in one; it responds to real-time screen context rather than delivering static scripts; and it is explicitly designed around learning outcomes, building understanding and independence, rather than just task completion.