The Agent-Ready Web: Building an API of Self
For two decades, the web has been optimized for human perception—layout stability, visual polish, and interactive refinement. In doing so, it has become increasingly opaque to the systems that will define the next layer of information consumption: autonomous agents. If a site is not machine-readable, it becomes a context void. Content is not simply harder to access—it becomes structurally unreachable, buried beneath rendering layers, JavaScript execution paths, and middleware abstractions that obscure semantic signal.
The Central Thesis: Emitting a Source of Truth
My work on tumaini.dk is centered on a simple idea: emitting a self-owned source of truth. In today’s web, identity is spread across many systems that are not under your control, which leads to fragmented and sometimes inconsistent representations of the same underlying work. The alternative is to rebuild coherence at the source by defining a personal API of self, where work, ideas, and development are exposed as a structured, machine-readable stream that reflects what is actually there rather than how it is presented.
The site itself is not designed around agents or automated consumption. It remains primarily a personal lab for structuring thoughts, testing ideas, and evolving concepts over time. The machine-readable layer exists as a secondary effect of that process—a way of exposing what is already there in a cleaner form, not the reason the system exists.
This is not about SEO or engagement metrics. It is about structuring information so it can exist in a more direct form outside of its presentation layer. In that sense, the source becomes the most stable version of the work, and everything else is just interface.
The Minimal Protocol of Identity Exposure
The implementation is intentionally minimal and focused on predictable access rather than presentation.
- /llms.txt (Discovery Layer)
A simple plaintext manifest that describes the system and its contents. It acts as an entry point so agents can quickly determine whether deeper access is relevant. - /llms-full.txt (Payload Layer)
A full export of the content, stripped of presentation concerns such as CSS, navigation, and layout structure. What remains is the raw material itself.
Together, these form a basic interface between identity and machine-readable representation, though they are not the primary interface of the site.
Demonstrating Structure in Practice
tumaini.dk is the practical example of this approach. Early versions revealed a common issue in standard web architecture: systems built primarily for human browsing introduce friction for anything trying to access content in bulk or in structure. Rendering requirements, client-side execution, and layered abstractions make extraction inconsistent.
By adding dedicated plaintext endpoints, that friction is reduced. The system can still be experienced in its intended form, while also exposing a cleaner representation of the same material. The change is not about optimizing for machines—it is about reducing unnecessary complexity in how the underlying work is accessed.
The Architect of Context
This shift changes how to think about building systems. The main optimization target remains the browser and human experience, but there is a parallel awareness of how information persists beyond that interface. The goal is no longer just to publish content, but to define a clear and stable structure underneath it that can be referenced without depending on presentation.
In this framing, building becomes a way of structuring output so it remains coherent outside of its original interface. The constraint is not visibility, but clarity: whether the underlying ideas can stand on their own when separated from how they are displayed.