Actuators over vibe coding for non-technical builders

It's 1:30 PM on a Friday and we're passing the mic around the room. I am sitting with builders researching emotional state in AI, a lawyer looking at knowledge graphs, performance marketers, and hardware design engineers.

Yao Kai Jiang dials in from the US. It's 10:30 PM for him. He is the founder of Momen, a full stack application builder for non-technical founders.

He frames the recent surge in prompt-to-app building as modality translation. He sees it as the process of translating human intent into formal software languages. Even with a perfect interpretation of your intent, an agent needs clear constraints. An agent operating without them just spins in a complete vacuum.

The problem with the spray gun

I have been watching the shift toward vibe coding workflows. Many builders use a pipeline from V0 to Cursor to Supabase to get an initial prototype up quickly. Yao Kai points out the hidden cost of this spray and pray method. Large language models are trained on massive amounts of human code. They gravitate toward what they know. That usually means frontend code. To further compound this phenomena, when a non-technical builder expresses intent, they also gravitate to visual effects.

The result is that a vibe coded app becomes some-what of a fat client: Logic gets duplicated across the frontend and makes the runtime brittle. The code itself remains completely invisible to the builder and just operates as a black box. Iteration becomes a serious problem as soon as the complexity of the project grows.

Re-platforming eventually becomes your only viable option. That puts a hard lid on your business traction.

Constraining the search space

Constraints serve a structural purpose. They reduce the branching factor at every step of a complex problem. Momen uses a visual canvas and a domain specific language to enforce these constraints.

But here’s the catch: AI agents struggle to write proprietary domain specific languages. They know Python and JavaScript well. But when you ask it to write Bubble code, you often end up in an endless loop of errors.

Yao Kai solves this with what he terms the actuator. An actuator packages rules and fixed procedures. In his view, the agent must operate the visual editor just like a human would instead of generating code in a vacuum. It runs type checks. It gathers execution logs. It relies on fast feedback to correct its path. The auditability stays on the exact same plane as the human operator.

Running real businesses

This framework translates to real scale. Yao Kai shares the screen of a sports card collection app. The founder is a solo non-technical builder. The app has around sixty thousand users and handles group buys and logistics. The builder manages about eight hundred thousand dollars in revenue a year.

Another builder runs a bar referral app in Hong Kong with twenty two thousand users. These are highly niche products.

To him, this is the niche that Momen serves. A dedicated software company would never serve a market this small because the engineering costs would never make sense. A solo builder with the right tools can own the market completely.

The boundary of control

Faye is a design engineer sitting across from me. She asks about AI coaching. She wants to know if platforms like Moment help non-technical founders slowly learn coding. Yao Kai is clear on the tradeoff. Platforms like Scratch optimise for deep understanding. Moment optimises for efficiency and control to help you ship a product.

Hanyi asks how actuators differ from Claude Code Skills. Yao Kai explains that actuators run inside a closed system. They initiate logic internally to maintain tight control over the platform state.

I ask him about how he’d position Momen against Supabase or Vercel. Yao Kai gravitates toward Supabase because he believes scalability comes from the backend. But he disagrees with the way Supabase forces dichotomy between edge functions and database functions. “The logic should live close to your data if your data comes from a database.”, he argues.

We finish the session and people stay to talk about their own stacks. The conversation shifts to database choices and vendor lock in. The tension between speed and auditability remains a constant problem for solo builders.

---

We run Co Working Fridays every week for builders in Singapore. This room had design engineers, lawyers, founders, marketers, and people building AI workflows across finance, manufacturing, research, and sales. The questions moved from whether non-technical founders should learn code, to how actuators compare with Claude Code Skills, to why the backend matters more once real data starts to persist inside a product people depend on.

Missed out last week? Don't worry, these conversations happen every Friday at SQ Collective.

Usually over laptops. Sometimes over pizza.

You're welcome to join the next one.