Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Green SARC, a governance framework that integrates financial and environmental cost controls directly into the architecture of agentic AI systems. By analyzing 3,000 real multi-step plans, the authors reveal that actual cost accretion surpasses linear predictions, while their architectural gate maintains a zero breach rate under binding budgets. The findings highlight the potential for substantial token, USD, and carbon savings, contingent on policy choices rather than mere architectural adjustments.
Real plans for agentic AI systems accumulate costs faster than predicted, revealing a critical gap in current governance frameworks.
Agentic AI systems act through tools and sub-agents, yet the controls meant to bound their financial and environmental cost still sit on dashboards evaluated beside or after execution. Green SARC applies the SARC governance-by-architecture framework -- four enforcement sites in the agent loop -- to FinOps and GreenOps, contributing the theory of what to enforce and how to predict it. We report four policy-independent results. (i) The unconstrained "State Snowball" is $Θ(n^2)$ in loop depth; on 3,000 real multi-step plans (SWE-rebench) it holds on 100%, with median curvature $\hat{c}_2=216$ exceeding the linear-accretion prediction $p/2=134$ -- real plans accrete faster than the model. (ii) On real residuals the Normal-$σ$ gate under-covers (92% at nominal 95%); split-conformal calibration holds (95.2%). (iii) A soft Lagrangian penalty tuned to the budget in expectation breaches it on 91.5% of seeds; the architectural gate breaches 0%. (iv) Under binding budgets the gate's over-budget incidence is 0% on synthetic and real (BurstGPT) arrivals. End-to-end token/USD/carbon savings (47--55%) are real but policy-dependent in magnitude -- set by a scope-cap knob, not by gate rejections. The library is open-source, dependency-free, and ships a regeneration script for every cited number.