By Bharath Yadla
SVP & Strategic Advisor, OCEO, Workato
Two distinct incidents clearly illustrate the scale and risk of Enterprise AI: Air Canada and Zillow.
Jake Moffatt chatted with Air Canada’s chatbot after losing his grandmother. The chatbot informed him that he could claim a discount after traveling, which contradicted the actual policy. When he attempted to claim it, the airline refused, claiming it wasn't responsible for the chatbot's advice. A lawsuit followed, and the tribunal ruled that a company is accountable for its chatbot's output (Moffat v. Air Canada, 2024). Ultimately, a probabilistic AI output became a legally binding commitment that the business had to own.
Similarly, Zillow relied on its algorithmic forecast to scale its home-buying business. Due to price unpredictability, it was forced to wind down operations, write off hundreds of millions, and incur a
$500 million loss while contractually bound to finish purchases. Deploying AI in core business operations requires strict controls and guardrails - it is not inherently foolproof.
Reasoning & Intelligence Commoditization
The capability horizon for task-frontier agents has roughly doubled every 7 months since 2019 (Kwa et al., 2025). As models rapidly commoditize, value shifts toward context and orchestration (Ankit, 2026). The Forrester State of Agentic AI 2026 report highlights a growing gap between chasing AI and capturing value, mainly due to enterprise trust. As reasoning advances, enterprises must implement governance and controls to ensure reliable operations.
When intelligence becomes abundant, the primary challenge is deploying Agentic AI in a governed manner to deliver reliable business outcomes.
AI is Stochastic; Enterprise Business is Deterministic
AI is inherently stochastic, while business operations are inherently deterministic. Despite huge momentum, enterprises struggle to extract value: an MIT study shows only ~5% of AI pilots reach production (Challapally et al., 2025), and Gartner reports only 13% of CIOs see transformational value (Woolcock & Brown, 2026).
AI relies on probability and prediction, introducing risks like hallucinations (Holtzman et al., 2020; Kalai et al., 2025). Conversely, business demands predictability. Payroll must strictly comply with wage laws, and purchase orders must be exact. Bridging this gap requires closing the “Governed Execution” gap.
Enterprise AI That Scales: Closing the Governed Execution GapScaling enterprise AI without costly failures requires mastering two critical pillars: execution and control.Execution is the workflow path converting an agent’s probabilistic decision into a deterministic business action across systems, processes, and people.Control provides decisioning authority over what an agent can access and execute, governing finance, privacy, scope, roles, and policy compliance.Gartner phrases it as the execution authority of an enterprise AI system, with agents operating with agent identity, policies, cost controls, execution scope, and explainability. Forrester (Joseph L, Dec 2025) indicates governance outside both the build and orchestration planes, so a controller must be neutral.LLMs/ Models can effectively & efficiently produce the reasoning & intelligence abilities across the business operations of an enterprise; however, deciding what actions to be executed and how they are governed has to be done deterministically, outside the boundary of LLMs/Models.Let me explain this concept through the lens of an aviation analogy: a trifecta that operates in coherence to keep passenger & flight safety.
Two Stories, One Boundary
Air Canada and Zillow failed not from bad intent, but from lacking governed execution over their models. Model outcomes must be treated as inputs to decisions, while execution and controls remain strictly managed.
To become an agentic enterprise, organizations should adopt a “Transformation First, AI Second” philosophy—focusing on transforming the business using AI rather than applying AI for its own sake (Yadla, 2026).
Reasoning is abundant. Control and execution decide who can use AI safely. Enterprises that scale AI will close the Governed Execution Gap.