It was the middle of the night, and I was re-translating an artwork statement into English to send to an overseas gallery. The creative work itself had gone fine — but the work of getting the piece out into the world was quietly consuming as much time as making it. I wanted to put a name to that feeling, so for the past three months I tested Antigravity as a production assistant alongside my creative work.
I'm an indie developer building apps on my own while maintaining an art practice that shows internationally. Carrying both solo means time for either is always limited. So the first rule I set was that Antigravity would never make art for me. My resistance to AI participating in the creative act itself hasn't changed after three months of use.
With that line drawn, what does the tool actually offer? Here's an honest account of what I delegated, what I didn't, and the reasoning behind each.
What I Expected from Antigravity (and What I Didn't)
What I was looking for was efficiency in the work that surrounds creation:
- Translating and polishing Japanese artist statements into natural English
- Coding exhibition pages and portfolio sites
- Building simple backends for app-form artworks
- Supporting correspondence with overseas galleries and curators
None of this is the creative work itself. All of it is what you have to do to get the work out into the world. Deciding in advance where that boundary sits — one step short of concept, composition, and color — meant I never had to wonder, in the moment, what was safe to hand over.
What Changed Over Three Months
English-language communication became much less costly
Artist statements, gallery inquiry emails, exhibition catalog descriptions — the psychological friction of writing in English dropped considerably.
What makes this work is asking for tonal calibration rather than translation. When I hand over a Japanese text, I fix the shape of the request like this:
Render the following artwork statement in English. Three conditions:
1. Not a literal translation — a register an English-speaking
viewer reads as native.
2. For words like "prayer," "negative space," "collective
psychology," don't flatten them into easy generics — offer two
nuance-preserving candidates each.
3. Where you've over-translated, leave a note so I can choose.
Ask simply to "translate this" and you get safe, flat prose. State where I want to decide, and the model shifts into offering candidates instead. I adjust for cultural nuance and make the final call on phrasing. What used to take half a day now takes 30 to 40 minutes, and that reclaimed time has gone back into the studio.
One caution: Antigravity produces text that is grammatically clean and conventionally structured. If the text itself is part of how the work communicates — if rhythm and weight carry meaning — the final precision is still yours to find. Antigravity gets you to competent quickly. Getting from competent to exact is still a human task.
Exhibition pages come together faster
Each exhibition needs a web page — images, statements, specifications. I used to build each from scratch. Now I have Antigravity generate a structural draft and make the visual decisions on top of it.
<!-- Example initial layout generated by Antigravity -->
<section class="artwork-display">
<figure>
<img src="/artwork/2026-series-01.jpg" alt="Artwork title" />
<figcaption>
<h2>Artwork Title</h2>
<p class="year">2026</p>
<p class="medium">Oil on canvas, washi collage</p>
<p class="size">H900mm × W1200mm</p>
</figcaption>
</figure>
</section>The moment I saw this draft, I knew: the caption should sit to the right of the image, the title wants a Japanese typeface, and I need substantially more negative space. None of these were ideas I had before seeing the draft — they emerged in reaction to it.
So now I hand over "the three things I will always override" up front: caption placement, typeface, and the proportion of negative space. Stating the constraints first raises the quality of the scaffold and cuts down the back-and-forth.
Production file organization became more systematic
Running app development and the art practice in parallel means production files accumulate without structure. I had Antigravity write a script to rename artwork image files by exhibition, medium, and sequence.
#!/bin/bash
# File organization script (generated by Antigravity, adapted for my workflow)
# e.g.: DSC_1234.jpg → 2026-solo-01-portrait.jpg
YEAR="2026"
EXHIBITION="solo"
SEQ=1
for file in ./raw/*.jpg; do
filename=$(basename "$file" .jpg)
newname="${YEAR}-${EXHIBITION}-$(printf '%02d' $SEQ)-${filename}.jpg"
cp "$file" "./organized/${newname}"
SEQ=$((SEQ + 1))
done
echo "✅ Organized $((SEQ - 1)) artwork files"For repetitive processing like this, Antigravity is fast and dependable. This is where I've seen the most straightforward return: manual tasks, now scripted, freeing attention for work that actually needs it.
What I Delegated, and What I Didn't
After three months the dividing line has come into focus. Laid out as a table, the basis for each decision is easier to see.
| Task | Delegated to Antigravity? | Reason |
|---|---|---|
| English polishing of statements | Yes (final call is mine) | Candidate drafting is fast; only nuance needs deciding |
| Exhibition page scaffolding | Yes (visual decisions are mine) | A draft becomes the trigger for my own judgment |
| File organization and renaming | Fully delegated | Repetitive work; easy to recover if wrong |
| Concept, composition, color | Never delegated | Comes from somewhere with no prompt interface |
| Final decisions about negative space | Not delegated | Emptiness is intention, not leftover space |
I draw this boundary with a single question: can I summarize the task in one spoken sentence? If I can, it's usually safe to delegate. The things I can't fully summarize — the use of negative space, the choice of color — are the ones that stay in my hands.
Negative Space Requires a Decision Not to Fill It
Of everything I kept for myself, negative space is the hardest to put into words. Left alone, Antigravity's layouts treat empty areas as space to be filled.
Even in the exhibition page above, the first draft crowded the image and caption together. I changed it to deliberately leave nearly a third of the vertical space open, so a viewer takes a breath before meeting the work — giving the emptiness a role of its own. I still haven't found a way to convey this "don't fill it" judgment through a prompt. So for negative space alone, I always re-open whatever was generated and decide it with my own eyes.
What Didn't Change
The work itself. I never once asked Antigravity what kind of artwork I should make.
Concept, composition, the specific choice of color — these come from somewhere that doesn't have a prompt interface. My practice took shape slowly, across years of moving between app development and showing work internationally. It isn't something a language model can generate, and I wouldn't want it to try.
If you find yourself asking an AI what your next creative project should be, that's worth examining — not because AI can't offer interesting provocations, but because the creative direction itself is where your specific irreplaceability lives.
Where Antigravity Is Strong and Where It Falls Short
Where it's genuinely strong
- Repetitive processing at volume: Batch renaming, metadata tagging, image-resizing scripts. Fast and dependable.
- Draft scaffolding: Getting something concrete on the page to react to, rather than building from blank.
- Consistency checking across documents: Keeping a series of English statements tonally unified across multiple exhibitions.
Where it falls short
- Series-level continuity: It processes pieces in isolation; it doesn't perceive how one artwork answers the one you made three years ago.
- Negative space as intention: As above — it optimizes for completeness in a way that works against restraint.
- Emotional precision in language: Logic holds, but the weight of a sentence and the quality of a pause still require human judgment at every point.
What Three Months Actually Taught Me
Month one: I looked for what Antigravity could do. Month two: what it couldn't do became clearer. Month three: I understood that the more clearly I define where my own judgment is required, the more useful the tool becomes everywhere else.
Vague prompts produce statistically safe outputs. Specific constraints and stated intentions produce genuinely useful responses within them. Working with Antigravity regularly is, in part, practice in articulating exactly what you want.
The same "delegate the periphery" instinct carries straight over to the development side of my work, where I run the Dolice Labs sites. I've written about layering installation production with a development workflow in a workflow for building art installations with Antigravity, if the boundary between making and coding interests you.
Where to Start If You Work Creatively
If your practice involves making things that require your specific, irreplaceable judgment, start with the work that surrounds the creative act rather than the act itself. Production correspondence, file organization, exhibition pages. When Antigravity takes those off the table, you get time back — and you can put it where it actually matters.
I'll keep refining this division of labor. For anyone working through similar questions in their own practice, I hope this account is useful.