At Ingenia we do not treat AI as a magic button or a substitute for judgment. We treat it as layers of the same flow: explore, design, prototype, produce motion, implement, and maintain. That means choosing tools with clear roles instead of stacking subscriptions nobody uses together.
How the pieces fit
- Figma Design and Figma Make for visual system, exploration, and functional prototypes before writing code.
- Higgsfield for generative video and image when the brief needs motion or fast campaign variations.
- Cursor and skill-based agents for implementation, review, and repetitive repo work.
- Hermes Agent when you need persistent memory, messaging channels, and orchestration outside the IDE.
- Next.js, Gemini, and RAG in client-facing product — like this site's assistant — when AI is part of the delivery.
The point is not «use AI» but not breaking the chain: what is decided in brand should reach the prototype; what is validated in prototype should have implementation criteria; what is generated in video should respect typography, rhythm, and tone already defined.
A good ecosystem reduces friction between phases. A bad one multiplies revisions and orphan files.
What we check before adding another tool
- Does it solve a real bottleneck or just add another interface?
- Does it export or integrate with what we already have (design tokens, repo, assets)?
- Can we audit and correct the output without depending on today's prompt?
A project from start to finish with this flow
To understand how the pieces chain together, think of a typical brand and web project. It starts by exploring: organizing proposition, audience and tone into a short direction, leaning on assistants to gather references and clear up ambiguity. Then the visual system is designed and the key screens are tested. Once the structure is validated, it is prototyped with real interaction so client and team can see how it behaves. If the campaign calls for movement, pieces are explored in parallel. Finally, it is implemented in code with review and tests, and the operation that keeps everything running is set up.
- Direction: proposition, audience and tone, in writing and in a few pages.
- Visual system: typography, color, components and rules of use.
- Prototype: key flows with interaction to validate before building.
- Motion, if needed: approved pieces that respect the system already defined.
- Implementation: maintainable, reviewed and tested code.
- Operation: follow-up, maintenance and automations that sustain what is published.
Where AI does not enter, on purpose
There are decisions we do not delegate: what to promise a client, what to tell about a case and what not, when a proposal is ready to send, what is published under the studio's name. AI speeds up preparation, exploration and repetitive tasks; responsibility for what goes out under our signature remains with people. That boundary does not limit the tool: it is what allows it to be used with confidence.
We also decide not to use it where it adds nothing. A text that has to sound like a specific client is worked on with them. An information architecture decision is thought through with paper and conversation before asking a model anything. And a piece that does not need motion does not get it just because the tool exists.
What stays when a tool changes
AI tools change quickly: alternatives appear, prices and terms are modified, a feature is deprecated. That is why we try to make sure what is valuable does not live inside any one of them. The design system is in our own files, the code in a repository of our own, brand decisions in documents anyone can read and working conventions written down. If tomorrow a tool stops suiting us, we replace it without losing the accumulated judgment.
Upcoming editorial posts go deeper on Hermes, Figma Make, and Higgsfield with the same lens: what they are for, where they help, and where not to use them.