The cost of building software has collapsed. What used to require a venture-backed team and a multi-million dollar seed round now requires one operator, a clear thesis, and a well-designed system. The bottleneck is no longer the code—it is the judgment required to architect the system and the discipline to operate it.
I am currently building Inky, an AI storytelling app. It is a core property within the Total Ventures portfolio. This is not a project meant for a quick exit; it is a digital product built to keep. Inky represents a shift in how we approach software: moving away from human-heavy development toward agentic engineering.
The Shift in Software Production
When building an AI story app today, you are not just writing functions. You are designing an environment where agents can execute work. In the old model, a developer wrote every logic gate. In the new model, I architect the operating system, and the AI workforce executes the production.
This shift changes the stakes. If you bolt AI onto an old organizational shape, you will be out-shipped by smaller teams with better systems. At Total Ventures, the goal is to maintain a portfolio of real products with zero employees. The machine is the moat.
Moving Beyond the Wrapper
Most people think building an AI story app is about finding the right prompt. I learned the hard way that a single prompt is a fragile foundation. If your product is just a thin layer over a large language model, you do not own a business—you own a feature that will eventually be absorbed by the platform providers.
To build for permanent equity, you must move beyond the wrapper. You need an orchestration layer that manages state, handles non-deterministic outputs, and ensures a consistent user experience.
The Orchestration Layer
Inky relies on a multi-stage agentic pipeline. Instead of asking a model to "write a story," the system breaks the process into discrete, manageable tasks:
- World Building: An agent defines the constraints, tone, and setting.
- Character Architecture: A separate agent develops character arcs and motivations.
- Narrative Synthesis: The engine weaves these elements into a coherent structure.
- Validation: A final agent checks for logical inconsistencies or tone shifts.
This is agentic engineering in practice. By decoupling these steps, I can swap out models or refine specific parts of the process without breaking the entire application. It allows for a level of precision that a single prompt can never achieve.
