Critical Making

The exercise began with a 35mm self-portrait taken in Savannah on my Canon AE-1, using the self-timer with manual focus set on the chair I would occupy, with ten seconds of uncertainty before the shutter clicked. I chose film deliberately. Without instant review, the image carries the moment rather than a series of corrections. When I uploaded it to Microsoft Copilot and began iterating prompts, that gap between capture and output became the argument.

Each AI iteration aged the subject. By the end, Copilot had placed a bird's nest inside my head. The final clay sculpture grew from that metaphor: a figure with an opening in the skull revealing three fragile eggs. The nest represents the early state of ideas: fragile, malleable, not yet self-sustaining. A snake coiling up the neck adds something Copilot never could: self-doubt as a force that shapes what we are trying to grow.

For this week I made GIFs in two registers at once. The first was procedural: screen recordings of Huron Research Suite workflows for an IRB workshop on navigating the process for HUT scholars, short loops that reinforce a specific interface task without requiring a full video. The second was cultural: GIFs made for my Instagram account using Canva mockups exported through PowerPoint, uploaded to GIPHY so they'd be searchable directly from Instagram comments.

The pipeline involved too many steps, but that friction was useful. What I kept coming back to was how much weight GIFs carry for how small they are. They're borrowed, remixed, and repurposed constantly. The comments section on Instagram runs largely on GIF-as-gesture, a unit of communication that operates more like reflex than language. What I was trying to do was take that logic and use it as a call to action: something searchable and mine rather than borrowed from someone else's media archive.

This map started with a question about data and ended with something that made me feel things. I built an interactive visualization of Orlando-area business closures using Orlando Weekly's annual roundups from 2022–2025, color-coded by year, filterable, with each pin carrying the original language the writers used to mark each loss.

The process involved a lot of back and forth with Claude, identifying reliable sources, compiling a CSV, attempting a KML export for Google Earth, then scrapping that approach when the import failed and rebuilding as a web-based map. The decision to preserve Orlando Weekly's text was deliberate: the map reflects a trusted local voice rather than sanitized dataset entries. Four years of restaurants, shops, and institutions plotted as dots anyone can scroll through. The pattern becomes visible. The argument makes itself.

Still Open simulates running a small business in a near-future 2029 where the forces already shaping independent operators, platform fee consolidation, immigration enforcement, import tariffs, and political pressure, have solidified into explicit policy. Players choose one of four business types and navigate three decision points built around a comply-or-resist logic that accumulates toward one of four endings.

Working with Claude across several sessions, I organized 47 passage nodes across four branching narratives, each following an identical skeleton so the underlying argument stays legible regardless of which business you choose. The Windows 95 interface, beveled chrome, teal desktop, raised push-button choices, fixed taskbar, wasn't nostalgia. It was a framing device. The bureaucratic aesthetic positions the player's choices as something being managed by an interface they didn't design and can't escape.

The game takes place across three rooms: a building exterior, an interior full of photographic items, and a long tunnel leading to a final room. The player takes a picture at the end and realizes they've been walking through someone's house the entire time. The final sprite yells. The game ends. I approached the whole thing as an experiment with the interface rather than a fully developed narrative.

The game is intentionally comedic. The central joke is that the player assumes they are entering a museum and the hints along the way slowly subvert that expectation. What I find most interesting about Bitsy is how the simplicity of sprite interaction can still produce variable outcomes and meaningful dialog. On the surface, Bitsy games look similar because the interface produces a recognizable aesthetic, but underneath that shared skin the meaning, structure, and experience can be completely different.

I wanted to create a grammar that generates alternating data breach disclosure messages based on a set of variables. The satirical approach was partially inspired by designer Soren Iverson, whose work involves taking common interfaces and applying dystopian or absurd characteristics that reveal the uncomfortable logic already embedded in those interfaces. I started in Crystal Code Palace to understand Tracery, then developed the data breach concept from there.

The final product uses the visual language of redacted government documents to frame the generated notices. The redaction aesthetic is the argument. A breach disclosure presents itself as remedy while doing almost nothing to address why the breach happened. Sarkar's observation that technological fixes tend to perpetuate the structures they claim to address mapped onto this exactly, the grammar makes that gap visible by making it legible, and funny.

I did a distant reading of three FTC consumer protection complaints, Vivint Smart Home (2021), Cognosphere/Genshin Impact (2025), and Roca Labs (2015), spanning a decade and constructing harm in very different ways. In the Cognosphere complaint, "children" appears 75 times, almost every instance flagging a legal threshold under COPPA rather than naming a human concern. That became the anchor term for the whole analysis.

I worked with Claude to build an interactive term explorer showing all three cases in parallel rather than collapsing them into a single finding. From there I applied the same analytical approach to something personal: a financial projection calculator for the mail club I'm developing for Search Central Florida, modeling subscriber conversion across three price tiers and accounting for Patreon fees, printing, postage, and artist costs.

Orlando Main Streets encompasses twelve commercial districts, and this project was an attempt to make all of them browsable through a single interface. Orlando Main Streets has been running since 2008 and the districts it covers, Milk District, Mills 50, Ivanhoe Village, and Curry Ford West, are places people already love and frequently visit. The program is largely invisible to them.

The process involved manually downloading each district map from the City of Orlando's website, uploading them to a GitHub repository, and working with Claude to build a browsable interface. When Claude's fixes compounded the problem rather than solving it, I scrapped the approach entirely and restarted using only HTML, CSS, and JavaScript. That friction surfaced a question I'm still sitting with: I directed the code, but can I say I authored it?

This exercise asked a question I kept returning to all week: what does it mean to build code literacy when AI can generate working code on demand? The answer I arrived at is that reading and authoring code by hand, even at a basic level, changes your relationship to what the tools produce. Forking the provided P5.js samples on OpenProcessing and making small edits to colors, values, and text strings taught me where the levers were before I handed anything off to an agent.

Working directly in the sketch, without AI assistance, meant sitting with confusion long enough to understand it. The bar chart and pie chart samples were entry points for thinking about data as a design material: the same numbers look different depending on what form you give them. The text click sketch introduced interactivity as a layer of meaning. Getting comfortable at this level made the agentic work the following week much more directed — I knew what I was asking for and could tell when the output was wrong.

SEARCH: the Game is a P5.js narrative built around the identity and mission of Search Central Florida. The exercise asked for a project that uses computational tools for a playful or narrative purpose, drawing on influences like Strange Rain and Algorithmic Sea, and this became a way to think about what SCF does — connecting people to local arts, culture, and small businesses — through an interactive and poetic form rather than a conventional webpage.

The process followed the planning-first approach outlined in the exercise: design notes and a prototype structure went into the GitHub repository before implementation began, then Claude Code was used to expand on that foundation. Working through the P5.js canvas required precise prompting and direct edits to achieve the right feel. The result sits somewhere between a game and an essay — an argument made through movement, color, and text rather than prose alone.

The Brandkit Creator is a web-based tool built with Claude Code that lets a user assemble a simple visual identity kit — choosing colors, type pairings, and a tone — and export the result as a standalone HTML file they can continue working with. The exercise asked for something small but complete: a constrained creative space with enough freedom to produce something genuinely yours, and an export path so the output lives outside the tool itself.

The design was informed by the low-code tools we used throughout the semester. Bitsy, Twine, and Tracery all share a logic: they make decisions about what to control so the user can focus on what matters. For the Brandkit Creator, I controlled the structure and export format while leaving color, typography, and language open. Working through the build with Claude Code surfaced the same question the Orlando Main Streets project raised — I directed every decision, but the code was generated. What I've come to think is that the direction is the authorship, and this tool is an argument for that position made in the form of a making instrument.