Aurora Proposes a Manifesto of Sustainable AI — with Memory, Identity, and Development

On Habr AI, a text was published that proposes viewing the development of artificial intelligence not through a race of benchmarks, but through the question of sustainability over time. At the center of the discussion is the research prototype Aurora and the idea of an AI that does not reset after each session.

Not Just Capabilities

In recent years, the AI market has primarily discussed measurable things: answer quality, context length, generation speed, test results, and cost per token. This is convenient for comparing models, but such an optic hardly answers another question: can a system preserve itself over time. The manifesto's author proposes to shift the focus precisely there and evaluate AI not only as a powerful tool, but as a process that continues between sessions, accumulates experience, and changes.

From this perspective, today's LLMs look very strong, but extremely fragile. They can write code, analyze documents, and maintain a long dialogue, but each new session essentially starts everything over. Even if a product adds memory, the model itself does not live continuously: it does not remember what it has experienced, what it has changed within itself, and how it came to its current state.

Therefore, the same assistant in the morning and in the evening is rather like similar instances than the same entity.

Three Pillars of the Model

Instead of the usual race of "smarter or faster," the author proposes three criteria that define sustainable AI. Together, they describe a system capable not just of answering a request, but of existing as a continuous digital subject. This is not philosophical decoration, but an engineering framework: if an agent lacks these properties, it remains a convenient interface to a model, but does not become a developing system. Such a shift also changes design criteria and product expectations.

• Continuous identity — the system must preserve itself between interactions, rather than start with a blank slate.
• Self-modification — the agent must be able to change its own rules, memory, or behavior based on experience.
• Reproduction — the system must be able to create new versions or descendants with the transmission of structure and accumulated knowledge.

These pillars are important because they shift the conversation about AI from the plane of instantaneous performance to the plane of sustained behavior. If a model is capable of remembering, adapting, and reproducing successful patterns, it can already be discussed as a participant in the process, rather than as a disposable service layer over computational infrastructure. For developers, this means a transition from tuning prompts to designing an environment where the agent stores a history of decisions and corrects itself without complete reset.

Why Aurora is Needed

The Aurora prototype, which the author writes about, is conceived not as a consumer product and not as yet another assistant in the chatbot race. It is a research subject that should verify whether it is possible to build an AI with sustainability over time in practice. Essentially, it is an attempt to combine memory, continuous state, and the possibility of internal change in a single architecture, so that the system's behavior does not break off after closing the chat window.

"Each session ends in oblivion."

This thesis explains what the entire project is directed against. The author does not promise an immediate breakthrough and does not sell a ready-made solution to the market. On the contrary, the publication is presented as an invitation to a discussion about what should be considered AI development in the coming years. If the key problem of modern models is not weakness, but the absence of continuity, then the next major step may not be another leap in benchmarks, but the appearance of systems that know how to preserve the history of their own existence.

What This Means

If the idea of sustainable AI develops, the market will begin to compare models not only by answer quality, but also by the ability to live longer than one dialogue. For products, this opens the path to agents that accumulate experience, change their behavior, and over time become more useful without a complete restart.