The enterprise world is currently grappling with a persistent void: the Impact Gap.
The Impact Gap is the widening distance between your digital promise, especially in the area of AI, and your business reality. Over the last few years, organizations have poured millions into AI pilots, yet most remain trapped in pilot purgatory—a state of high activity and significant spend that fails to manifest as measurable ROI.
The reason? We are trying to force probabilistic reasoning systems to do the job of deterministic systems without a bridge. To move from a slow, expensive reasoning pilot to a fast, profitable autonomous enterprise, you need a way to turn highly contextualized logic into reflexive habits without resorting to rigid and brittle code filled with an unending maze of if-then-else loops nested one inside the other.
Like all transformational problems, the path to achieving enterprise autonomy is not a light-switch event. Rather, it’s a carefully constructed journey. This journey is mapped out in Boomi’s Agentic Blueprint, our foundational artifact that serves as the operating system for enterprise autonomy. The blueprint describes the path across four layers of capability, starting with Layer 1: Trusted Data & Systems and culminating in Layer 4: Goal Realization & Digital Impact. Within Layer 4, we define the anchor metric for autonomy: the Decision Delegation Index (DDI)—or what we colloquially call “The Leash.”
The Mental Model: Instinct vs. Reasoning
To understand the difference between traditional deterministic IT systems and the more flexible systems possible with AI, it’s helpful to turn to a classic work of psychology that sheds new light on how we ourselves go about thinking and solving problems.
In his seminal work Thinking, Fast and Slow, Daniel Kahneman defines two modes of human thought:
- System 1 (The Instinctive Layer): Fast, automatic, and deterministic. This is where a process fires predictably every single time without conscious effort.
- System 2 (The Reasoning Layer): Slow, analytical, and effortful. This is where an AI agent thinks through a complex, non-linear problem.
As Matt McLarty recently explored, this cognitive architecture provides the perfect blueprint for agentic automation. Traditional automation (ERP, ETL, etc.) is a System 1 environment—it’s “hardened” logic, proven, and reliable. AI agents, by their nature, live in System 2—they are “reasoning” assets, better at working through novel or unpredictable problems.
Now let’s go back to the Impact Gap. The Impact Gap persists because enterprises are trying to scale System 2 reasoning as if it were System 1 instinct. In other words, smitten with the promise of AI, they’re trying to build and scale AI systems as though they were traditional deterministic IT systems designed to solve rote, predictable problems. Just as in real life, applying the wrong mode of thinking can backfire — imagine trying to make a snap decision about a highly complex medical diagnosis. In the same way, applying System 2’s AI reasoning to System 1 automation yields poor results.
Fortunately, there’s a way forward out of this misapplication of technologies which results in the Impact Gap. We call the process of bridging this gap between goals and results Decision Hardening.
The Milestones of Agentic Impact: You Can’t Skip Steps
The biggest mistake leaders make is attempting to jump straight from Human-Only processes to Autonomous + Accountable systems. In Boomi’s Agentic Blueprint, we define this journey through the Decision Delegation Index (DDI, which we’ll explore in more detail in the next blog post in this series). The DDI is a measure of how much you can trust an agent to reliably produce trustworthy results. The higher the trust, the more leeway you can give an agent. We colloquially call this granting of leeway “The Leash.” (Dog owners will appreciate this metaphor. The more you can trust your pooch to behave in public, the longer the leash you can give it. As with dogs, so with AI agents.)
You cannot lengthen the leash until you have proven the reliability of an agent. There are no shortcuts. The more you can trust agents, the more tasks, workflows, and roles you can give them to automate.
The progression of granting more and more trust follows a strict evolutionary path:
Human Only: A task context where humans perform tasks with no form of delegation to agents.
Human-in-the-Loop (HITL): A model where a human must review and approve every agent action before execution.
Human-on-the-Loop (HOTL): A model where the agent acts autonomously, and the human audits results by exception.
Autonomous & Accountable: A state where AI-based agents are trusted to make decisions and take actions based on those decisions. Just as with humans in the workplace, Boomi’s Agentic Blueprint requires transparency and accountability for agents to be given autonomy. When scaled enterprises give free rein to agents without ensuring transparency or accountability, harsh consequences are sure to follow.
Each of these 4 stages are represented in Boomi’s Agentic Blueprint as milestones (summarized in the table below). As your enterprise matures in your ability to enable autonomy safely, increasingly complex and valuable use cases become unlocked for your teams to enable with agentic solutions.
| Milestone | Oversight Model | DDI (The Leash) | Role of the Human |
| Agent Incubation | Human Only | 0.0 | Performs 100% of the task; no delegation. |
| Guided Action | Human-in-the-Loop (HITL) | 0.1 – 0.3 | Validates agent suggestions; clicks “Go” on every action. |
| Optimized Scale | Human-on-the-Loop (HOTL) | 0.4 – 0.8 | Audits by exception; requires Pass@1 excellence. |
| Autonomous Enterprise | Accountable Autonomy | 0.9 – 1.0 | Sets goals and monitors intent; system is self-governing. |
Why the “Skip” Fails
Attempting to skip from the Agent Incubation or Guided Action phase to an Autonomous Enterprise without a hardening phase is effectively running with scissors. Since most AI business cases rely on high-scale efficiency, a snapped leash (a predictable reaction to a spiking Policy Violation Rate (PVR)) usually means the project will fail to meet financial targets given that the cost of sustained human labor wasn’t part of the plan — making it an immediate candidate for defunding that will push you back into Pilot Purgatory.
To close the gap, enterprises must align on a concept of strategic maturity for decision hardening, where strategic maturity is only achieved when your Delegation (DDI) increases while your Violations (PVR) remain at background-noise levels.
The Bridge: The Reliability Gate
To safely lengthen the leash, you need a Reliability Gate. Boomi’s Agentic Blueprint leverages Boomi’s Agentic Evaluation Framework to create this gate. This methodology uses metrics pioneered by OpenAI and Anthropic — specifically Pass@k and Pass^k — to provide the empirical evidence needed to move an agent from the Reasoning of System 2 to the Instinct of System 1.
You can’t skip the steps, but with the right measurement framework, you can climb them much, much faster.
In our next blog post, The Science of Trust — Moving Beyond Vanity Metrics with Pass@k, we’ll explore using these new agentic metrics to create the Reliability Gate you need for safely moving your agentic automation from one phase to the next.