Entering a New Domain in the Era of Generative AI: An Autoethnography of Solo Game Development as a Bioinformatics Educator

Synthesis: Over 73 days in the summer of 2026, a bioinformatics educator with no game-development experience built a create-your-own-adventure game about bioinformatics and kept a public journal of the attempt. Kim's autoethnography analyzes 25 entries totaling 14,844 words, written between May 24 and August 5, 2026, and finds that Generative AI made the domain enterable — Anthropic's Fable 5 built a retro game from a zero-shot prompt in less than 10 minutes — while the identity work moved at a different speed. Communities of practice, not the model, were the primary source of information, and they were also where the practitioner learned that open GenAI use carried stigma in game development. Prior expertise did unexpected work: data-structures knowledge reframed the branching design, and evolutionary biology supplied the lens for reading claims about artificial neural networks. The paper's claim is not that the Learner Agency GenAI conferred was illusory, but that being enabled to enter a domain is not the same as belonging to it, and belonging is what formed the new identity.

Key Findings

1. Prototyping made entry feel feasible in minutes. Anthropic's Fable 5 built a retro game from a zero-shot prompt in less than 10 minutes, reframing game development as something a newcomer could attempt.
2. Communities of practice, not the model, were the main information source. The autoethnographer mined the social platforms it lists and a local indie playtest event, treating community norms as accumulated heuristics.
3. Prior knowledge shaped both the game and the reading of GenAI. Bioinformatics expertise produced a dystopian premise, and data-structures reasoning about a depth-10 branching game ruled out a perfect binary tree.
4. GenAI use was selective and self-assessed. The practitioner delegated planning, storywriting, prototyping, and image processing but not playing games, reading community posts, or interrogating creativity — activities they enjoyed or felt competent at.
5. Game-development communities did not openly acknowledge GenAI. At a playtest event GenAI was completely absent from discussion, and social media carried fierce rejections, pushing the design toward conservative systems and Creative Commons Zero resources.
6. The budget was a consumer subscription, and it moved. The practitioner held a $0 computing budget and a daily free quota of five queries for vision models, then raised the ceiling to no more than $20 per month.
7. Identity formation ran at a human pace. Seventy-three days produced a game and a beginner's grasp of the domain, but the paper argues that a new identity took longer than the capability that prompted the attempt.

What the journal recorded

The study is a first-person account of one practitioner entering an unfamiliar domain with an explicit educational goal: a create-your-own-adventure game in which a player uses a bioinformatics tool and learns to interpret its output. Data collection ran from May 24, 2026 to August 5, 2026 and produced twenty-five journal entries totaling 14,844 words, an average of 593.76 words per entry (minimum 291, maximum 1,223), and ten supporting images. The journal sat on a site built with Lovable; ChatGPT and Claude assisted with editing, Figma and the Computer Vision Annotation Tool produced non-AI images, ChatGPT and Adobe Firefly generated some AI images, and ChatGPT, Cursor, and Claude supported planning and testing. GenAI use was documented for every occurrence, and analysis followed the six-step reflexive thematic analysis described by Braun and colleagues. Two questions organized the coding: how a practitioner engages with a new domain, and how prior knowledge is integrated with new domain knowledge. The journal remains publicly readable, which is unusual for experiential accounts of this kind.

Belonging to communities of practice

The practitioner's engagement followed a three-step loop the paper names explicitly: acquiring new information, interrogating relevant knowledge, and updating priorities. Direct experience was occasional and mostly GenAI-assisted; indirect experience was frequent, formative, and came from communities rather than from the model. The autoethnographer attended a playtest event hosted by a local indie game developer community, barely knowing more than a dictionary definition of the word "playtest", and learned the community's own measures of success — being accepted to a gaming convention, being featured in a magazine, releasing a game on Steam. What struck them was the silence about GenAI: no developer mentioned it, and social media showed open hostility, so the practitioner leaned toward more conservative systems and Creative Commons Zero resources. Adjacent communities were more receptive — interaction designers shared prompting practices, software developers compared API costs against running a local system, health sciences educators discussed how to teach proper use, and biomedical researchers asked for evidence of performance. The paper reads the gap as an asymmetry between the pace of technical development and the pace of adoption, and borrows the Ship of Theseus to ask whether an identity survives when its defining activities are replaced by different processes with comparable outputs. The timeline it reproduces dates vibe coding to February 2025 and harness engineering to February 2026, and records that practitioners stayed with the latter.

Prior knowledge as design material

Bioinformatics did more than supply content. Because the practitioner had expertise in evolutionary biology and pandemic-related research, the game acquired a dystopian shape: a protagonist who had survived twelve consecutive pandemics and was the only person eligible to analyze the collected samples. The story was, in the practitioner's phrasing, "a suspicious future", since bioinformatics analysis is potentially more susceptible to automation than wet-lab analysis. The same expertise worked as an error detector — falsehoods were immediately visible when underinformed GenAI drafted story material — and as an analytic instrument. Reasoning about branch points from data structures, the practitioner saw that a deeper educational game tree would demand more storywriting than one person could sustain, and turned to more feasible structures. A zoo visit produced a sharper result: comparing AI companies' biomimicry of artificial neurons to tiger ocelli, which mimic the function of eyes without the morphology, made the metaphor distasteful while the performance remained impressive. The educator identity contributed a different set of judgments. The practitioner found mixed evidence that gamification improves performance but substantial evidence that it improves engagement, and noticed that learners in memorization-intensive fields showed declining Motivation as GenAI demonstrated strength at memorization. Those commitments shaped the branching design toward a skewed tree with a clear overarching goal.

What this means for practice

• Instructors. Structure domain-entry assignments so that joining a community is part of the task, not an optional extra: the practitioner's information and design standards came from a playtest event and the platforms they followed, and the model alone would not have supplied them.
• Faculty developers. Teach the acquisition–interrogation–prioritization loop as an explicit routine for staff entering an unfamiliar field, and require written reflection on what prior expertise is doing to the new domain, since the practitioner's disciplinary lens both helped and biased the design.
• Researchers. Read single-case autoethnography as evidence about process, not effect, and pair it with designs that can separate community participation from tool access as the driver of identity change.

Limitations

• The design is one autoethnographer over 73 days and 25 journal entries, with no comparison case and no second coder reported; the autoethnographer is also the analyst.
• Data collection ran through summer, when community gatherings are rarer; the paper notes that more frequent person-to-person engagement might have been achieved with a longer period.
• Person-to-person contact amounted to one playtest event plus professional communities that were extensions of existing connections, so claims about community norms rest on a thin slice of direct contact.
• The practitioner was not a full-time game developer and capped GenAI spending at $20 per month, with a daily free quota of five queries for vision models early on, which bounds what could be tested.
• Capability claims are tied to models available in summer 2026, including Anthropic's Fable 5; the paper names no version for the ChatGPT and Claude instances used, and no task-level outcome measure was collected.

Citation

Kim, H. (2026). Entering a New Domain in the Era of Generative AI: An Autoethnography of Solo Game Development as a Bioinformatics Educator. arXiv:2611.09623 [cs.HC].