Substrate-Aware AI Agents: Execution Context as a First-Class Input

2609.05232 ↗ · 2026-09-04 · cs.AI

Manu Agrawal

8 pages, 3 figures. Reproducibility artifacts and source-linked evaluation code: https://github.com/manu2/Context-Aware-Agent-Experiment

The finding

Headline claim

Telling an LLM the execution environment's RAM and time limits before it writes code (substrate awareness) makes it proactively choose lower-memory, faster implementations, without any algorithmic hint or repair loop.

The test

Single-turn numerical code generation: an 8,000 x 1,024 float32 pairwise Euclidean-distance task given to three frontier configurations (Claude Opus 5, GPT-5.6-Sol, Gemini 3.7 Flash), five samples per condition, with versus without a 128 MB / 10 s contract, plus a 96 MB extension. Outcomes: correctness, peak RSS (MaxRSS), wall time on a pinned local Python 3.9 environment.

Effect size & scope

Contract disclosure lowered peak memory in 13 of 14 executable paired comparisons and cut mean wall time 1.68x to 3.09x across cohorts. At 96 MB, correct-and-within-budget outcomes were 4/5 (Claude), 5/5 (GPT), 3/5 (Gemini) versus 0/5, 1/5, 0/5 task-only; mean MaxRSS fell 49-74% and wall time 35-64%. Effect demonstrated on exactly one task, one runtime, tiny n=5 per condition.

Limitations

Authors concede cohorts are independently sampled, not statistically matched; configurations are diverse rather than tier-matched; results are descriptive with no significance testing; one task and one deployment environment only; the broader agenda (runtime versions, GPUs, tools, quotas) is untested. Obvious unstated limits: n=5 per cell, single deterministic micro-benchmark, measured MaxRSS may not reflect cloud billing or production workloads, and results may not generalize beyond numerical NumPy code.

Counterpoint

This is a proof of concept at toy scale: one synthetic task, five samples per condition, no error bars, no baseline of simply asking models to write efficient code (efficiency-focused prompting without limits), and possible prompt-sensitivity, a different phrasing of the constraint could change outcomes. A single-author arXiv paper with a hand-built benchmark could reflect prompt-engineering rather than a general substrate-awareness principle, and 'up to 3.1x faster' headlines the best cohort.

Economy implication

Little or nothing yet economically. If the principle generalizes, agent harnesses that pass deployment contracts into planning context could cut cloud compute waste and first-pass deployment failures in agentic coding products, but this paper alone does not establish that at production scale.

Who this matters for

Technical Capability: Agent builders get controlled evidence that injecting execution constraints into the planning context changes first-pass implementation choices, a cheap, training-free intervention worth testing in real harnesses.

Market & Valuation: Cloud and agent-platform vendors have a small early signal that context-exposing features (contract injection) could differentiate agent runtimes by reducing over-provisioned compute and failed deployments.

Abstract

Autonomous AI agents increasingly select actions in environments whose memory, execution-time, runtime, compute, and operational constraints determine what counts as a suitable plan. We call the absence of this execution context from an agent's planning state substrate blindness. We test this general proposition through numerical code generation, where selected implementation choices and operational consequences are directly observable. Three frontier model configurations--Anthropic Claude Opus 5, OpenAI GPT-5.6-Sol, and Google Gemini 3.7 Flash--generate code for a high-dimensional pairwise Euclidean-distance task either from the task alone or with a 128 MB RAM and 10.0 s wall-time contract. Contract disclosure reduced measured peak process memory in 13 of 14 executable index-aligned task-only versus contract-disclosed comparisons and reduced mean wall time in all three cohorts, making execution up to 3.1x faster. Across the audited corpus, disclosure produced structural code changes including bounded blocking, float32 retention, upper-triangle traversal, and in-place or memory-mapped buffers. At a tighter 96 MB contract, independently sampled contract-disclosed cohorts achieved correct-and-within-budget outcomes of 4/5 for Claude Opus 5, 5/5 for GPT-5.6-Sol, and 3/5 for Gemini 3.7 Flash, compared with task-only outcomes of 0/5, 1/5, and 0/5; cohort mean MaxRSS and wall time were 49-74% and 35-64% lower than their task-only references. These results establish a controlled proof of concept for substrate-aware agent planning: a minimal execution contract induces proactive structural adaptation in generated programs, shifting computation away from unconstrained allocations and substantially improving observed resource-time profiles before execution.