AI Agents Need Decision Rights
Agents cannot become safe enterprise actors until organizations define decision rights, escalation paths, evidence requirements, human sign-off, logging, and outcome accountability.
Executive Summary
The brief in four points.
- 01
Agentic AI shifts the question from what AI can generate to what AI is allowed to do.
- 02
Deloitte’s operating model themes emphasize decision rights, escalation, rationale documentation, logging, human sign-off, and outcome measurement.
- 03
Gartner has warned that many agentic AI projects may be scrapped because of cost, maturity, and unclear business value.
- 04
Large People Model interprets agentic AI as a Decision Architecture and Governance Architecture problem before it is a tooling problem.
Why This Matters Now
The market shift is operational, not just technical.
Agents are different from assistants. Assistants usually wait for prompts. Agents can pursue objectives, coordinate tasks, call tools, move information, and trigger workflows. That means agents participate in the operating model.
As soon as an agent can act, the organization needs to know what decisions the agent can make, which decisions it can recommend, what evidence it must produce, when it must escalate, and who owns the outcome.
Without those boundaries, agentic AI creates automation without accountability. The system may move faster, but leaders may lose visibility into why actions occurred and who is responsible for their consequences.
“Agents cannot scale safely without decision rights, escalation paths, and human accountability.”
Market Signals
Signals executives should not ignore.
40%+
Agentic AI projects Gartner has warned may be canceled by 2027
Cost, maturity, unclear business value, and weak risk controls are common failure conditions for agentic AI.
Source: Gartner coverage50%
Agentic AI projects reported as stuck in proof of concept
Pilot-stage friction often reflects orchestration, governance, data, and production-readiness issues.
Source: Dynatrace coverage77%
Leaders saying AI adoption is outpacing governance capability
Agentic AI increases the gap between what systems can do and what governance can supervise.
Source: IBM survey coverageSignal Pattern
Deloitte’s operating model work for humans and governed agents highlights the need for decision rights, escalation, evidence, human sign-off, logging, rationale documentation, and outcome measurement.
Gartner has projected that a significant percentage of agentic AI projects may be scrapped because of cost, maturity, and unclear business value.
The market signal is straightforward: autonomy increases the value of decision architecture and governance architecture.
LPM Interpretation
How the seven-layer model explains the pattern.
Large People Model places this issue in Decision Architecture and Governance Architecture. Decision Architecture defines what choices exist, who owns them, how evidence is used, and when decisions are revisited. Governance Architecture defines the boundaries that keep decisions safe, compliant, and aligned.
Agents also touch Communication Architecture because their rationale must be communicated clearly. They touch Information Ecology because their recommendations depend on trusted context. They touch Platform Structure because their actions occur inside systems.
The agent question is therefore not simply “Can the agent do it?” The executive question is “Should the agent be allowed to do it, under what evidence standard, and with what human accountability?”
This layer is part of the operating model pattern the resource is diagnosing.
This layer is part of the operating model pattern the resource is diagnosing.
This layer is part of the operating model pattern the resource is diagnosing.
Leader Questions
What leaders should ask before scaling.
What decisions can agents make, recommend, or only prepare?
Which decisions require human sign-off before execution?
What rationale, evidence, and source context must be logged?
When should an agent escalate to a human owner?
How will outcomes be measured and revisited?
Visual Callout
Decision rights map
Agents need explicit boundaries between assist, recommend, decide, execute, and escalate.
Sources / References
Reference signals used for this brief.
Deloitte humans + agents operating model research
Reference theme: agentic operating models require decision rights, escalation paths, logging, rationale documentation, evidence, human sign-off, and outcome measurement.
Gartner agentic AI cancellation projection coverage
Reference theme: agentic AI projects may be canceled when cost, maturity, business value, or risk controls are unclear.
Soft Lapemo Connection
From framework to operating system.
Large People Model defines the framework. Lapemo is being built to help organizations operationalize it across ownership, decisions, information, governance, and governed agents.