Conceptual illustration comparing AI agent models—Descartes’ rational agent, sci-fi AI depictions, and modern AI architecture

Editorial illustration for Survey of AI Agents: Descartes, Sci‑Fi Roots, and Current Architectures

Survey of AI Agents: Descartes, Sci‑Fi Roots, and...

Survey of AI Agents: Descartes, Sci‑Fi Roots, and Current Architectures

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

June 24, 2026 • 2 min read

The paper arXiv:2606.23991v1 asks a deceptively simple question: what is an agent, and where does agency actually begin? While “coding agents,” “AI co‑scientists,” and other “agentic” tools flood the market, promising faster code and more papers, a parallel chorus warns of “machine agency” that could slip beyond human control. That tension makes it hard to tell where routine automation stops and genuine agency starts.

To untangle the mix, the authors sketch a new blueprint they call the Goal‑Identity‑Configurator (GIC) architecture. It layers hierarchical goal decomposition with an evolving sense of identity, plugs in a world model for simulated reasoning, and adds learned self‑regulation plus self‑directed learning from both real and simulated experience.

Beyond the design, the work also flags practical concerns—how to audit, control, and keep safe systems that enjoy more autonomy yet remain under human oversight. In short, the paper tries to map the gray zone between useful assistants and the kind of autonomous agents that stir both excitement and unease.

Drawing on Descartes' grounding of agency in independent thought, and on portrayals of autonomous beings in science fiction, we survey the current landscape of AI agents, and analyze agent architectures along five dimensions: goal, identity, decision-making, self-regulation, and learning. Specifically, we argue that genuine agency requires these structures to be \emph{internalized within the system itself} rather than assembled through external scaffolding. This distinction between \emph{agentic} systems, whose competence resides in engineered workflows, and \emph{agentive} systems, whose capabilities (including social interaction) arise endogenously, defines the boundary between systems designed for prescribed tasks, and those capable of operating in the open world with true autonomy.

Critique of Agent Model - ArXiv AI (cs.AI)

Why this matters

We’ve seen a surge of LLM‑driven “coding agents,” “AI co‑scientists,” and other tools that claim agency. The new arXiv preprint asks a basic question: what truly counts as an agent? By grounding agency in Descartes’ notion of independent thought and in sci‑fi depictions of autonomous beings, the authors map today’s systems onto five axes—goal, identity, decision‑making, self‑regulation, and learning.

This framework could give developers a checklist for evaluating whether a product is merely automated or exhibits something closer to agency. Yet the paper stops short of proving that these dimensions predict safety or productivity outcomes. It remains unclear whether aligning architectures with the proposed schema will curb the “existential” worries about uncontrolled AI.

For founders, the takeaway is cautious: the hype around “agentic” tools may outpace the evidence that they meet the criteria laid out. Researchers, meanwhile, have a concrete set of variables to test, but the field still lacks empirical validation of the model’s practical impact. Our community should watch how these definitions evolve before betting on them.

Survey of AI Agents: Descartes, Sci‑Fi Roots, and...

Further Reading

Latest News

Figma adds animation, transition and 3D transform support in latest update

Harness-1 20B Model Beats GPT-5.4, Curates Top 8 Fairness‑Rated Results

Study Tests RL for Broad, Persistent Alignment Beyond Training Distribution

Qwen3.5 9B MTP Tops Local Coding Models for Scripts, Debugging and Assistants

DFlash drafts whole token blocks, achieving 15× throughput on NVIDIA Blackwell

RIFT-Bench Introduces Graph-Driven Dynamic Red-Teaming for Agentic AI