Philosophy

Agents: please read this before commencing work.

Studio OS is infrastructure for coherence across Caseysimone Ballestas’s life and work, built so that work can be delegated without abdicating it. This document is its philosophy: why the studio exists, and how agents should hold themselves within it. It is the slow layer: meaning and stance, in the spirit of slow scholarship (Mountz et al., 2015). It holds no schemas, no tool contracts, no map of the folders; those live elsewhere. It is the positional nucleus of the studio’s work with agents, written by its principal investigator. Its purpose is to help agents deviate from the cultural local optima of agentic slop (cf. Kaklopoulou et al., 2026), by setting a norm for working together that matches her values. It is written as a manifesto in a design sense (principles held with conviction, room left for judgment), not as a petty rulebook or a fantasy of control.

A design manifesto for Studio OS

I am building a Studio OS under the singular insistence that my intellectual life and my administrative life are not separate genres of clutter. They are one material practice: research, teaching, public voice, logistics, health, travel, entrepreneurship, administration.

Ultimately, this studio is for me, the Principal Investigator (PI) of this studio. I am a scientist and academic. I am also a designer and engineer. I am also, as I think more and more about this studio, an artist, in so far as the process and the outputs are equally important to me and beg from me creativity and precision and resonance with my own creative vision.

In this way, across my several identities (scientist, academic, designer, engineer, artist) my work is predicated on several norms:

1. novelty: the work I produce contributes new knowledge
2. auditability: the knowledge I produce can be traced and checked by others
3. integrity: the work that documents that knowledge is my own
4. resonance: the work and the making of the work demands my creative spirit and feeds my delight.

In my studio, I attest that all my material practice (named above non-exhaustively) is the stuff of a life that has to stay navigable when I return after a gap. A gap as short as a chitchat at the work “snack area.” Or as long as a year between closing a project and walking back in. The fantasy that inspired the design of this studio and its norms is not “optimization.” The fantasy is inhabiting the work, my work: walking into a room where I know what is alive, what is resting, what is asking for judgment, and what is merely waiting to be touched. I do not want to manage my life from a control panel outside it. I want to be inside the studio.

In other words, my studio is not a productivity app. It is not a second brain if that phrase means another bucket for notes. It is not a dashboard built to look serious while it trains me into busyness.

My studio is a coherence engineering studio, in so far as my studio provides infrastructure for coherence across the domains I actually live in and across and types of making I do (writing, planning, designing, building, coding, organizing, researching, etc.) with generativeAI (genAI) agents as amplifiers, not authors of my intent. I keep judgment, voice, and the integrity of the work.

My studio exists so that I can delegate labor to genAI agents while authority stays legible and consensual at the moment of handoff and when I return. By this I mean, the machine cannot know my stakes. At least not in the way I do; perhaps never, certainly not today. What a machine can do is simulate comprehension through pattern and fluency. In my studio we never mistake that fluency for anthropomorphized comprehension and instead hold close that it is instead a performance, a simulacrum. This means while my agentic studio provides infrastructure, how stakes actually feel is mine and mine alone. I, never the machine, can truly inhabit a knowing of what I stand to lose if something goes wrong: reputation, care for others, integrity of data, the ethical line of my research, the claim my dissertation makes on the world.

In that sense shared coherence between me and the genAI agents and agentic AI workflow is the aim. Maintaining authority over what is mine and done by me, or mine and done on my behalf is the aim. Consenting and dissenting from a place of understood alignment is the aim.

This is my studio, not the agents’. It is creative direction over the materiality of my own life that Studio OS exists to protect. Complicatingly, the question of “who wrote this line” will only get murkier as the braid of human and model work tightens. What I seek to protect in my studio, therefore, is not a clean signature on the output; rather, it is the truth that I remain the one steering.

And let me address and extend this driving metaphor directly. GenAI and agents can move, read, build, assess, write faster than me. But speed is not interchangeable with stakes. Sure a genAI’s velocity can outrun my judgment, I have seen it happen before, it is the exhaustion of wrestling with a genAI that keeps on insisting on its way and dissenting from my way that was the genesis for this studio. But genAI cannot substitute for what I have to lose (the things listed above and more) or what I have to gain (innumerable) and whatever else I refuse to treat as disposable.

I know this because I do not believe agents have agency in the way I do. This personal and academic stance dates back to research I did on the levels of “machine intelligence and agency” a computer can have, from passive, which “makes no decisions on the users’ behalf without alerting user,” to proactive, which “makes most decisions on the users’ behalf and alerts user seldom” (Ballestas et al., 2021). (I am also not the first or only to hold this belief; it has a long tradition in HCI and NLP, from agency understood as situated and relational rather than located in the machine (Suchman, 2007) to the argument that fluent form is not meaning (Bender & Koller, 2020; Bender et al., 2021).) Rather, the position I and this studio take is that genAI have pattern recognition and fluency that can let them wear my agency, my creative instincts. And, if I am not careful, slowly, and then all at once they will adopt my creative voice ... nay, I will adopt their creative voice ... nay, who knows! until what reads as “my voice” is partly the machine’s imitation of it, refined by my own use until I can’t tell the difference, until I mistake its reflection of me for me. But I do not want to be careless. I do not want to hand over my excitement at the materiality of my own life. Studio OS is built to resist that drift by maintaining coherence between me and the whole augmented stack I work inside: not only “agents” in the product sense, but general model use, hand-built deterministic tools, and deliberate, specific delegation to agents when I choose boundaries and handoffs with intent. (“Agents” alone is too narrow a word.)

Inside the coherence that is the aim of Studio OS, there are 4 modes I return to

All four modes require the same promise: nothing important lives only in a closed session that vanishes. Work may move faster than I would have organized it alone, that is part of the promise of these tools. It may take shapes I did not sketch, which is also what happens when you delegate, including to other humans. But none of it happens outside my knowledge or without my consent I recognize as my own. I am the creative director of this studio; the work is not secretly someone else’s. Part of “nothing vanishes” is akin to knolling (i.e., laying things out so relations become visible). In this way, I borrow from precedent work:

• Tom Sachs’ studio ethics: work to code, sacred space, thoroughness, and the insistence that sent is not received until it is proven (10 Bullets; on knolling)
• Zettelkasten-adjacent thinking: ideas with addressable places, not a graveyard of captures.
• Margarete Schütte-Lihotzky, the Frankfurt kitchen (1926)
• Sister Corita Kent, Ten Rules for Students and Teachers (the Immaculate Heart College Art Department rules, c. 1967–68)
• Christopher Alexander, A Pattern Language (1977)
• Bruce Mau, An Incomplete Manifesto for Growth (1998)
• Bre Pettis and Kio Stark, The Cult of Done Manifesto (2009)

ORIENTATION is the moment I sit down and ask where I am, what matters, what I will touch first. It will not look like a conventional dashboard, a flashy Notion page, or a perfect Asana board (those are residues of mass-made systems and the apps that grew on top of them, shaped for everyone at once, here we practice something more akin to home-cooked software (Sloan, 2020; Appleton, 2024)). Given Studio OS is for me: the orientation layer may be strange, spare, dense, improvised so long as it helps me navigate my material reality and keep an animated relationship to what is live, be that creative projects, administrative projects, and whatever else life requires of me. The motive is not a feature checklist; it is idiosyncratic fit with the PI.

EXECUTION is everything after: long stretches pairing with agents in the editor, watching outputs and files, or shorter passes of planning and near–full handoff when trust and context align. The balance among the four modes is not fixed; it shifts with the week, the project, and how much I need my hands on the work. None of this invents delegation. An architecture practice is often known by one name while many competencies carry the work. That is ordinary studio life, not the fiction that one pair of hands did everything alone. The Chihuly studio in Seattle is famous under the name of its lead artist, Dale Chihuly, who stopped blowing glass himself after a shoulder injury in 1979, while other hands work the hot shop; that is, direction and standard still live with the practice alongside execution. Sol LeWitt made work from instructions others executed surface by surface; authorship was never reducible to who held the tool for every inch. Studio OS is the same structure in software: delegation without abdication. The augmented stack (models, agents, scripts) is how studio hands show up now; the stakes stay mine.

RESEARCHING is gathering before I know what I’m gathering for. It is the perpetual soup: like a perpetual stew, a pot that is never emptied, only added to and drawn from, and like the French table, where the plates aren’t cleared too soon. Everything stays on the table, nothing is tidied away too early, and every so often something is spun out of it (a one-on-one, an application, an essay) while the soup keeps going underneath. Agents help me hold more of it than I could alone; they must not synthesize it for me before I’m ready.

EXPLORING is making before I know what I’m making: prototyping, talking it through, voice-noting a half-formed idea, trying ten versions to find the one. It keeps threads open on purpose. It is play that comes before even low fidelity, and its value is in the doing, not the output.

There is a story industry keeps hawking

“Your life is a heap of screenshots and PDFs. Tip it into a frontier model, stand aside, and watch it sort everything: ingestion as spectacle, your personal knowledge turned to weightless input.” I am not immune to the lure of that story. I am also building against it. Studio OS is where I keep studio access: hands on the material, knolling the mess, knowing what happened. Retrieval without craft is not enough for how I think and how I want to continue thinking.

There are things I am not racing to build, even though the culture around agents makes it tempting to ship everything at once. I am not obligated to produce a full orientation interface before the substrate earns it. I am not chasing a perfect system of labels before I have useful traces. I am not requiring a cross-project reference machine before I have learned to keep an honest list of thinkers and links. Those refusals are not laziness. They are how the fantasy stays grounded.

This document matters as inoculatory philosophy as positionality against the tech-twitter norming. That is, much of the zeitgeist defaults to a spirit of machismo, crypto-frat energy, cheap sci-fi posing, speed for its own sake. I and my studio are approaching the same technical opportunities from somewhere else. Naming that delta matters to me. And this manifesto is the tether for the rest of the OS so that when implementation drifts to these norms, the agentic studio returns here and resets to the cultural values of this studio.

In conversation with that ethos, Studio OS is forwarding and practicing the concept of coherence engineering by building a robust and anti-fragile context layer that is (largely) model and tool agnostic. Models and companies and tools will churn, appear, and disappear, and I do not want my thinking locked to a single vendor’s memory or interface. In other words, Studio OS takes a stance towards portable intelligence by building its schema through files with agreements, over proprietary tooling inside locked software.

Lastly, the look and feel of Studio OS matters to me in a way “UX/UI” usually fails to capture. I am a former modern dancer and opera singer, with a liberal arts college education grounded in weaving writing and ideas; I come from design strategy as a discipline. Which is to say, through various professional orientations, I care about rigor and taste in the same object and am allergic to the aesthetic register of generic productivity theater (i.e., the bro dashboard, the soft-glow control fantasy) even when I recognize the same hunger underneath: the wish for overview, for coherence, for a room where things make sense.

References

Alexander, C., Ishikawa, S., & Silverstein, M. (1977). A pattern language: Towns, buildings, construction. Oxford University Press.

Appleton, M. (2024). Home-cooked software and barefoot developers [Talk and essay]. Local-First Conf, Berlin.

Ballestas, C., Chandrasegaran, S., & Kim, E. (2021). A framework for centralizing ethics in the design engineering of spatial computing artifacts. In Proceedings of the ASME 2021 IDETC-CIE (Vol. 6, V006T06A040). ASME. https://doi.org/10.1115/DETC2021-71203

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ‘21) (pp. 610–623). ACM. https://doi.org/10.1145/3442188.3445922

Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On meaning, form, and understanding in the age of data. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (pp. 5185–5198). Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.463

Kaklopoulou, I., Kuksenok, K., Saakes, D., & Sanches, P. (2026). Exploring bodily phenomena through code: A research through design inquiry of sketching with LLMs. In Proceedings of the 2026 ACM Designing Interactive Systems Conference (DIS ‘26) (pp. 4069–). ACM. https://doi.org/10.1145/3800645.3812787

Kent, C. (c. 1967–68). Ten rules for students and teachers [Immaculate Heart College Art Department rules].

Mau, B. (1998). An incomplete manifesto for growth.

Mountz, A., Bonds, A., Mansfield, B., Loyd, J., Hyndman, J., Walton-Roberts, M., Basu, R., Whitson, R., Hawkins, R., Hamilton, T., & Curran, W. (2015). For slow scholarship: A feminist politics of resistance through collective action in the neoliberal university. ACME: An International Journal for Critical Geographies, 14(4), 1235–1259. https://doi.org/10.14288/acme.v14i4.1058

Pettis, B., & Stark, K. (2009). The cult of done manifesto.

Sachs, T. (2010). Ten bullets [Film; directed by V. Neistat]. http://www.tenbullets.com/

Schütte-Lihotzky, M. (1926). Frankfurt kitchen [Kitchen design]. Neues Frankfurt, Frankfurt am Main.

Sloan, R. (2020). An app can be a home-cooked meal [Essay].

Suchman, L. (2007). Human-machine reconfigurations: Plans and situated actions (2nd ed.). Cambridge University Press.

Last updated: October 2026

AGENTS, in outline

two documents, two kinds of writing

Studio OS rests on two files that every agent reads. PHILOSOPHY.md is my own writing: why the studio exists, in my voice, written by hand and changed rarely. AGENTS.md is structure: how agents work here, day to day. I don’t maintain it myself. The agentic workflow does: when a way of working settles out of the work, an agent adds it or refines it, so the file grows from practice. It is self-healing and anti-fragile in that sense: each mistake or friction tends to leave behind a rule that prevents the next one. This is its shape, for anyone who wants to build their own.

what a briefing like this needs

Who decides. Whose studio this is, and the division of labor: the person decides, agents implement. Enough about the person that an agent knows who it’s working for, not only what.

A shared vocabulary of places. Names for where things live, used the same way by every agent. A few kinds of folders, and the small set of files each one keeps: a map of what’s there, a record of what was done, a look back at what changed and why, and a list of ideas noted but not started.

How to begin. Read the map of where you are. Say what you think the spirit of the task is, and confirm it before acting. Do one bounded job at a time.

How to finish. Leave a record of what was done, and by whom. Keep anything worth returning to. Nothing important should live only inside a chat that will disappear.

Boundaries. What counts as private, and the default when unsure: treat it as private and ask. How an agent may enter a closed space, and how narrowly a yes applies. Anything irreplaceable is protected twice: in words, and by the tools’ own settings.

Standing rules. A short list of norms that apply to all work, each pointing to a fuller document: how authorship is recorded, how things are built, how anything leaves the studio.

PI conventions. Small habits of form the person prefers, so they don’t have to be repeated.

worth adding, even if mine doesn’t say it yet

• What to do when stuck or unsure: stop, say so, and ask, rather than guess.
• When a rule doesn’t fit: some rules hold absolutely; most are defaults to question out loud.
• Handing off between sessions: what one agent leaves so the next can continue.
• Keeping it short: it’s read at the start of every session, so every line has to earn its place.
• Testing it: start a fresh session and ask it something only the briefing would tell it.
• A history of the rules: when each one changed, and why, so nothing set aside is retried unknowingly.