tasoeur

PS: I did Tokyo quite a lot, Shanghai too, though you usually need to take the metro at some point. San Francisco is also great if you don’t mind the hills (and also use common sense).

Sadly as an indie (or frankly anyone these days?) if you don’t have a social media following, you simply don’t exist. Distribution is one of - if not the most - important things to have, so farming social media is unavoidable. The worst part is that you can’t just post and go, these apps (and respective algorithms) expect you to use them regularly for your content to be shown to people.

Automating all of that is also super tricky since there are a billion bots out there. And the alternative being pay-to-win with expensive advertising.

My summer project was to challenge existing 2D / 3D visual effect creation tools (touch designer, comfyUI, etc.) with something agentic as the true foundation (not just slapping MCP on top of an existing paradigm).

It also came from the following observations:

- people are making their own coding harness, why not one for realtime visual effects?

- using Claude code or codex to make art and installations is absolutely fantastic, but you usually end up with a lot of friction and paper cuts when it comes to deterministic workflows and concept iteration.

SubjectiveZero is the free and open source project that resulted from it. It’s still following the classic node editor paradigms but making the nodes absolutely more malleable while also combining a user experience flow optimized for ideation and iteration. I definitely invite you to check out the showcase and examples on the website, but basically it feels kind of magical when your editor just “get” what you want, and nudge your exploration into adding more little knobs here and there so using the tool feels more like playing rather than “operating”.

I also liked the emerging concept of an adjustable abstraction ladder, meaning people who are very technical can still get as much granular control as they wish to while people who just want high level concepts and vibes can just use prompts to try things out.

When it comes to determinism, the tool has some neat features like an agent graph to constrain agent workflows, overpowers MCP that drives user interface changes as the agent makes progress, allows the agents to verify their work and coordinate so the usual painful wait for agents to be done is somewhat minimized. There’s also a way to do basic model routing to tweak your intelligence/speed ratio.

I’ve personally used it to make art and fun experiments. Feel free to send feedback or contribute tokens to the project if the concept speak to you!

I feel like CLI would probably work too but then I’d end up implementing something similar. That being said, I’ve been bitten by the usual suspects: too many tools will cause context windows to grow quickly and some agents will sometimes skim through a subset of the tool list without querying the entire thing, causing incorrect behavior.

If you have tokens to burn I invite you to check out the source code see how extensively it’s being used: https://github.com/sxp-studio/subjective-zero

(video to see the MCP in action, it’s a bit long so feel free to skip: https://www.youtube.com/watch?v=DcI1tsPJ8eM)

Another kind of cool use of MCP that I’ve encountered is actually from… the French government! They do it for their open data initiative: https://github.com/datagouv/datagouv-mcp

Current project: https://sxp.studio/apps/subjectivezero

If you’re open to the idea of composing code blocks and ideas, plus some generative UI exploration, feel free to join!

- https://sxp.studio/apps/subz

Little plug since we’re in the topic of dumb phones etc, I have a completely free and non-creepy iPhone app launcher for folks who are into that:

- https://sxp.studio/apps/applist

(The app catalog is also open source)

Then… I wanted to apply this to (beginner level) language learning. And yep you have to be disciplined to go through the cards daily, but it works.

My app is 100% free so my trick was to have users write what they need, the app generates a prompt to give to any AI provider, which in turn gives you back a JSON that will be converted into a deck.

Nothing truly ground breaking, but again it works for beginner level concepts :-)

My first project was an “old school” node editor done almost entirely with agentic coding. I slapped an MCP on it to see if the agents could make cool stuff and it was mostly hit or miss, but nothing too mind blowing. Blog post about it: https://sxp.studio/blog/subjective-building-a-native-vfx-edi...

My second take on this is to channel coding agents into “prompt nodes” coupled with some agent orchestration. This helps define clear data flows and API contracts for each node + overall graph. This strategy works better since we’re constraining the problem space and overall getting better results when combined (and it also plays nice with context window limitations). The current demo is a bit slow and basic but it’s looking promising! https://sxp.studio/apps/subjectivezero

I wrote a blog post about my process: https://sxp.studio/blog/subjective-building-a-native-vfx-edi...

...and you can download the app here if you're curious (the app is free!): https://subjectivedesigner.com

Next project is going to be a pivot of that project into something related to creative coding and agentic :-)