The five categories of Muda, borrowed from lean manufacturing and applied to autonomous AI.
In lean manufacturing, Muda (waste) is any activity that consumes resources without creating value. Toyota identified seven categories of waste and built an entire production system around eliminating them. The result was not just efficiency, but quality. Less waste means fewer defects, shorter cycle times, and more predictable outcomes.
The same principle applies to AI agent operations. Every unnecessary permission, excess capability, and redundant data exposure is waste. Every unit of waste is an unpriced option on a security incident, a compliance violation, or an operational failure.
Corvair's patent defines exactly five categories of operational waste in AI agent systems. Each is computationally measurable and actionable.
Definition: Authority held but not required for the declared mission.
Permission Waste occurs when an agent has been granted access rights, API permissions, or system privileges that exceed what its current mission requires. This is the AI equivalent of giving every factory worker a master key to every room in the building.
Why it is dangerous: Every excess permission is an unpriced option on a security incident. If the agent is compromised, the attacker inherits all granted permissions (not just the ones the agent was using). Permission Waste directly inflates both Cumulative Operational Authority (COA) and Blast Radius.
How Corvair measures it: The governance engine compares the agent's granted permissions (from the Registry baseline) against its actual permission usage during runtime. The delta is Permission Waste.
How Corvair eliminates it: Zero Standing Privilege (ZSP). Agents never carry persistent credentials. Every action requires a just-in-time privilege request that is evaluated and scoped to the minimum necessary authority.
Definition: High-risk abilities the current mission does not require.
Capability Waste is subtler than Permission Waste. An agent may have been built with code execution capability, file system access, or external API integration, but its current mission is document summarisation. The unused capabilities carry latent risk even if they are never invoked.
Why it is dangerous: Capabilities represent potential actions an agent can take. Unlike permissions (which are externally granted), capabilities are intrinsic to the agent's design. An agent with code execution capability can be coerced into executing malicious code through prompt injection, even if it was deployed for a completely different purpose.
How Corvair measures it: The Agent Registry Layer 4 (Capabilities, Tools, and API Integrations) declares all capabilities. The governance engine compares declared capabilities against the mission scope (Layer 10) and flags any capability that exceeds mission requirements.
How Corvair eliminates it: The CI/CD Governance Gate blocks deployment of agents with capabilities that exceed their declared mission. Existing agents are assessed through the automated discovery pipeline.
Definition: Data access beyond strict operational need.
Exposure Waste occurs when an agent has access to data domains, knowledge bases, or information systems that its mission does not require. This is the data equivalent of Permission Waste, but focused specifically on information exposure.
Why it is dangerous: Every unnecessary data exposure increases the Blast Radius if the agent is compromised or behaves unexpectedly. In regulated industries, exposure to customer PII, financial records, or health data beyond operational need may itself constitute a compliance violation.
How Corvair measures it: The Agent Registry Layer 7 (Data Domains and Knowledge Sources) declares authorised data scopes. Runtime telemetry tracks actual data access patterns.
How Corvair eliminates it: Data access is scoped per-action through the just-in-time credential system. Agents receive ephemeral access tokens limited to the specific data domains required for the current action.
Definition: Unnecessary network hops introducing latency and systemic risk.
Transport Waste occurs when data moves through systems, networks, or environments that it does not need to traverse. This includes unnecessary API bridges, redundant data transformations, and multi-hop delegation chains where a direct path exists.
Why it is dangerous: Each additional network hop introduces latency, creates potential interception points, expands the attack surface, and adds systems to the compliance scope. Transport Waste is often invisible because it accumulates gradually as architectures evolve.
How Corvair measures it: The Agent Registry Layer 5 (Environment and Execution Context) defines the expected operational topology. Runtime telemetry tracks actual network paths and API call chains.
How Corvair eliminates it: The governance engine enforces environment and network constraints at the action level. Agents are constrained to declared execution contexts.
Definition: Historical error rates and policy violations indicating process fragility.
Defect Waste captures the operational cost of failures: policy violations, incorrect outputs, timeout errors, and any action that did not meet the defined quality standard. Unlike the other four categories, Defect Waste measures outcomes.
Why it is dangerous: Defect Waste directly indicates Process Sigma degradation. Rising defect rates signal that the agent's reliability is declining. Left unmeasured, defects compound through delegation chains and erode Agent Sigma.
How Corvair measures it: Every governance decision (grant, deny, downgrade, sandbox) is logged with causal explanation. Policy violations are counted, categorised, and tracked over time.
How Corvair eliminates it: The DMAIC feedback loop ensures that recurring defect patterns trigger design-time improvements rather than accumulating as technical debt.
| Waste Category | Measurement Method | Primary Reduction Mechanism |
|---|---|---|
| Permission Waste | Granted vs. used permissions | Zero Standing Privilege (ZSP) |
| Capability Waste | Declared vs. mission-required capabilities | CI/CD Governance Gate |
| Exposure Waste | Authorised vs. mission-required data scope | Per-action data scoping |
| Transport Waste | Expected vs. actual network topology | Environment constraints |
| Defect Waste | Policy violation rate over time | DMAIC feedback loop |
The risk of authority accumulation beyond operational need.
How systemic waste erodes AI performance and security posture.
Start measuring operational waste in your agentic systems. Our diagnostic tools quantify the delta between authority and mission.
Schedule a Briefing Download Framework