Steerability via constraints: a substrate for scalable oversight of coding agents
2026-07-02 • Artificial Intelligence
Artificial IntelligenceCryptography and SecuritySoftware Engineering
AI summaryⓘ
The authors explain that while coding agents (like AI programmers) can write code, human oversight is the main challenge and can be costly and risky if agents are unchecked. They suggest using traditional team management practices such as access control, strict coding rules, and network policies to guide these agents, which is more efficient than recent complex AI supervision methods. Their experiment with a Python project containing hidden backdoors showed that these controls significantly improved the detection of problems. They chose Python because it has fewer built-in safety features, but they believe these ideas can apply to more secure languages like Rust as well.
coding agentshuman oversightaccess controlnetwork policiescoding conventionscodebase securityPythonbackdoorssoftware scalabilityRust
Authors
Thomas Winninger
Abstract
Coding agents are capable; human oversight is the bottleneck. Unconstrained agents introduce security risks, erode codebase scalability, and make human review increasingly costly. We argue that the same methods used for decades to manage large human engineering teams: access control, network policies, strict coding conventions enforced by tooling; transfer directly to coding agents, and are cheaper (in token) than recent agentic scaffolding. We sketch a start-to-end system on this principle, and report a controlled experiment in scalable oversight: a small reviewer (Gemma 4 e4b) inspects a Python codebase containing 11 inserted backdoors. Recall rises from 54.5% (unconstrained, no tools) to 90.9% (constrained substrate plus a ~200-LoC `docs` CLI), with substrate and tools contributing independently. We choose Python deliberately: substrate-level oversight gains are largest where the language gives the fewest guarantees by default; the principles extend to languages like Rust.