Rethinking AI Hardware: A Three-Layer Cognitive Architecture for Autonomous Agents

Li Chen

cs.AIApr 15, 2026

Rethinking AI Hardware: A Three-Layer Cognitive Architecture for Autonomous Agents

Authors: Li Chen

Abstract

The next generation of autonomous AI systems will be constrained not only by model capability, but by how intelligence is structured across heterogeneous hardware. Current paradigms -- cloud-centric AI, on-device inference, and edge-cloud pipelines -- treat planning, reasoning, and execution as a monolithic process, leading to unnecessary latency, energy consumption, and fragmented behavioral continuity. We introduce the Tri-Spirit Architecture, a three-layer cognitive framework that decomposes intelligence into planning (Super Layer), reasoning (Agent Layer), and execution (Reflex Layer), each mapped to distinct compute substrates and coordinated via an asynchronous message bus. We formalize the system with a parameterized routing policy, a habit-compilation mechanism that promotes repeated reasoning paths into zero-inference execution policies, a convergent memory model, and explicit safety constraints. We evaluate the architecture in a reproducible simulation of 2000 synthetic tasks against cloud-centric and edge-only baselines. Tri-Spirit reduces mean task latency by 75.6 percent and energy consumption by 71.1 percent, while decreasing LLM invocations by 30 percent and enabling 77.6 percent offline task completion. These results suggest that cognitive decomposition, rather than model scaling alone, is a primary driver of system-level efficiency in AI hardware.

Figures & tables

Figure 1: Tri-Spirit Architecture. Intelligence is decomposed into planning ( $L_{s}$ ), reasoning ( $L_{a}$ ), and execution ( $L_{r}$ ), coordinated via the asynchronous SpiritBus $\mathcal{B}$ .

Figure 2: Task execution flow. $L_{a}$ parses user input, decomposes it into a task queue, and dispatches sub-tasks to $L_{r}$ for execution. Repeated task sequences may be promoted to habit policies via $\mathcal{H}$ , bypassing $L_{a}$ on future invocations.

Figure 3: Habit compilation pipeline. High-frequency task sequences detected by $L_{a}$ are abstracted into triggers and compiled into stateless policies deployed to $L_{r}$ , eliminating LLM inference for those task classes.

$\downarrow$ lower is better; $\uparrow$ higher is better. P95 latency: Tri-Spirit 2,274 ms, Cloud 2,474 ms, Edge 214 ms.
System	Mean Latency (ms) $\downarrow$	Energy (mJ) $\downarrow$	LLM Calls $\downarrow$	Offline (%) $\uparrow$
Cloud-Centric	2,146 [2,138, 2,153]	46.1 [45.9, 46.3]	1.00	0.0
Edge-Only	179 [178, 180]	11.8 [11.7, 11.9]	1.00	100.0
Tri-Spirit	523 [486, 562]	13.3 [12.5, 14.1]	0.70 [0.65, 0.74]	77.6

Table 1: Simulation results at default thresholds,

N=2{,}000

, seed = 42. Values are mean with bootstrap 95% CI in brackets. Best value per column in bold .

Figure 4: Main simulation results (seed = 42, bootstrap 95% CI shown in panels (b) and (c)). (a) Latency CDF: Tri-Spirit dominates Cloud-Centric across all percentiles; Edge-Only achieves the lowest latency via exclusive local execution. (b) Mean energy with bootstrap CI. (c) Mean latency with bootstrap CI.

Figure 5: Mean latency by task type. Type-A and Type-C show the greatest reductions under Tri-Spirit; Type-B improvement is modest because 55%+ of Type-B tasks are escalated to $L_{s}$ .

Figure 6: Sensitivity analysis. Shaded bands span results across five values of the complementary $\gamma$ parameter; solid lines correspond to default $\gamma$ values. (a) $\tau_{r}$ vs. mean latency; (b) $\tau_{a}$ vs. mean latency; (c) $\tau_{r}$ vs. energy (left axis) and offline continuity (right axis); (d) 2-D heatmap of mean latency over $\tau_{r}\times\tau_{a}$ grid, with the default configuration marked ( $\star$ ).

Variant	Lat. (ms) $\downarrow$	Energy (mJ) $\downarrow$	LLM calls $\downarrow$	Offline $\uparrow$
Cloud-Centric	2,146 [2,138, 2,153]	46.1	2.00	0.0%
Edge-Only	179 [178, 180]	11.8	1.00	100.0%
TS-LocalOnly	93 [89, 97]	6.2	0.50	100.0%
TS-RandomRoute	523 [485, 560]	13.2	0.67	77.6%
TS-NoReflex	601 [564, 639]	18.4	1.12	77.6%
TS-NoHabit	533 [493, 571]	13.9	0.72	77.6%

Table 2: Ablation results (

N=2{,}000

, seed = 42, shared noise). CI = bootstrap 95%. Best value per column bold .

Figure 7: Ablation results. (a) Mean latency with 95% CI across all seven variants. (b) Energy per task. (c) LLM invocations per task. (d) Latency savings decomposition: each bar shows the absolute latency reduction attributable to each component; TS-Full net latency shown for reference.

Note: components are not strictly additive due to interaction effects; the sum is computed end-to-end.
Component	$\Delta$ Latency (ms)	% of Cloud Gap
Local execution (avoid cloud)	$-$ 2,053	95.7%
Reflex layer (fast reactive path)	$-$ 78	3.6%
Habit compilation (zero-inference)	$-$ 10	0.5%
Intelligent routing (quality alignment only)	$\approx$ 0	$<$ 0.1%
Total (Cloud $\to$ TS-Full)	$-$ 1,623	75.6%

Table 3: Latency saving attributed to each component (relative to cloud-centric baseline, 2,146 ms).

Edge-Cloud Collaborative Inference

See all 26 papers

[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI
Apr 15, 2026You Rim Choi, Subeom Park, Hyung-Sin KimAdaptive Sensing Embodied AI
Seoul National University, Seoul, South Korea
The Auton Agentic AI Framework
Feb 27, 2026Sheng Cao, Zhao Chang, Chang Li +3Agentic AI LLM Agent Orchestration
Networking-Aware Energy Efficiency in Agentic AI Inference: A Survey
Apr 9, 2026Xiaojing Chen, Haiqi Yu, Wei Ni +5Energy-Efficient ML LLM Inference Efficiency
Shanghai University, Shanghai, China · Nanyang Technological University, Singapore · Edith Cowan University, Perth, Australia +2

Apr 15, 2026J/K move · Enter open · S save

[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI

You Rim Choi, Subeom Park, Hyung-Sin Kim

Seoul National University, Seoul, South Korea

Adaptive Sensing Embodied AI Cognitive Architectures for AI Agents Active Perception Hardware-Software Co-Design Edge Inference

Open paper AskPDF arXiv

As AI moves from data centers to robots and wearables, scaling ever-larger models becomes insufficient. Physical AI operates under tight latency, energy, privacy, and reliability constraints, and its performance depends not only on model capacity but also on how signals are acquired through controllable sensors in dynamic environments. We present Artificial Tripartite Intelligence (ATI), a bio-inspired, sensor-first architectural contract for physical AI. ATI is tripartite at the systems level: a Brainstem (L1) provides reflexive safety and signal-integrity control, a Cerebellum (L2) performs continuous sensor calibration, and a Cerebral Inference Subsystem spanning L3/L4 supports routine skill selection and execution, coordination, and deep reasoning. This modular organization allows sensor control, adaptive sensing, edge-cloud execution, and foundation model reasoning to co-evolve within one closed-loop architecture, while keeping time-critical sensing and control on device and invoking higher-level inference only when needed. We instantiate ATI in a mobile camera prototype under dynamic lighting and motion. In our routed evaluation (L3-L4 split inference), compared to the default auto-exposure setting, ATI (L1/L2 adaptive sensing) improves end-to-end accuracy from 53.8% to 88% while reducing remote L4 invocations by 43.3%. These results show the value of co-designing sensing and inference for embodied AI.

Abstract

Edge-Cloud Collaborative Inference

Explore similar work

[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI

The Auton Agentic AI Framework

Networking-Aware Energy Efficiency in Agentic AI Inference: A Survey