Considering agent-web interactions and the future of the web

The broader implication is that the web may evolve toward becoming natively agent-readable, enabling more powerful, predictable, and secure autonomous systems.

Schultze, S., Kietzmann, M.V., Schonfeld, N., & Stock-Homburg, R.M. (2025). Building the Web for Agents: A Declarative Framework for Agent-Web Interaction. arXiv. https://arxiv.org/pdf/2511.11287

The paper suggests a future where we may effectively see two parallel webs emerge: the human-facing web we use today and a machine-readable agent web built on explicit affordances. By introducing VOIX, a declarative HTML framework that exposes structured state and actionable tools directly to agents, the authors argue that websites can reduce the need for brittle inference and improve both safety and performance. Their results show that agents operating on these explicit affordances execute tasks far more reliably and with significantly lower latency than those trying to reverse-engineer the DOM or interpret screenshots. If this approach gains momentum, we could see a bifurcation in the ecosystem where traditional interfaces continue to serve human users while a new, parallel layer of agent-facing markup enables precise, predictable interaction for autonomous systems. So much of the current internet is not going to get rewritten with VOIX this is like drawing a line in the sand for the before times and the future.