In my previous research on Model-Driven Software Development (MDSD), we argued that MDSD and the REST architectural style form a synergistic relationship. As software teams move toward AI-driven development, that architectural fit is becoming even more relevant.

The core argument still stands: a truly RESTful system, driven by hypermedia, provides the abstraction and decoupling needed for scalable software. What changes now is that AI makes several long-standing hypermedia benefits far easier to realize in practice.

1. Formal Constraints as a Foundation for Generative AI

REST is defined by strict architectural constraints. In an MDSD context, these constraints enabled a high degree of automated code generation. For Large Language Models (LLMs), the same strictness acts as a guiding framework by defining how interfaces should be structured and how interactions should evolve.

2. Managing Hypermedia Complexity through Automation

Developing and maintaining comprehensive hypermedia-driven APIs (HATEOAS) has historically been considered expensive. MDSD showed that this complexity can be handled through formal models and generators. AI now offers a more adaptive route to the same outcome by reducing manual boilerplate and supporting the generation and refinement of self-descriptive interfaces.

3. Bridging the Gap to Intelligent Clients

The main criticism of hypermedia-driven systems was the “last mile” problem: clients still required human-directed logic to interpret and act on hypermedia controls. With modern AI agents, this limitation weakens. Agents can process affordances and relation types in real time, navigate application state transitions dynamically, and reduce reliance on hard-coded URI knowledge.

Why return to a Hypermedia-Driven architecture now?

By re-embracing the original hypermedia constraint (HATEOAS) in an AI-driven ecosystem, we can realize several architectural advantages:

• Software longevity: interfaces support independent evolution and long-term stability.
• Enhanced decoupling: client and server communicate through standardized media and relation types.
• Reduced client complexity: application state transitions are driven by server-provided controls.
• Agentic discoverability: APIs become naturally compatible with autonomous agents that navigate by discovery instead of fixed paths.

Hypermedia is dead, long live Hypermedia.

If you want to take a look at my paper: https://conferences.ulbsibiu.ro/icdd/2022/files/ICDD2022_Proceedings.pdf