Business intelligence, an essential component of every enterprise organization, can be obtained fairly effortlessly by the evolution of modern Artificial Intelligence (AI) capabilities. But reaching this point has taken time. Over more than three decades, intelligence passed through four distinct eras to arrive at today’s dawn of agentic analytics, which is the strongest capability we’ve seen thus far.
Agentic analytics will reward companies that take the time to train AI on their business — how it’s structured, what its terminology sounds like, and how its software connects to form a system. Agents can investigate questions, weigh options, and recommend decisions, and their answers earn trust only when shared definitions, governed data, and open standards sit underneath them.
The process of implementing agentic AI is similar to onboarding a new employee — albeit one that wades deep in the details. Here’s more on how agentic AI is coming to be, where the technology is headed, and what business leaders can do to prepare for the future.
How has business intelligence evolved?
The rise of business intelligence began with IT. Under this system, data specialists would build reports whenever a question was asked, and everyone else had to remain captive to IT’s time constraints. The system worked, even if it was quite laborious and involved plenty of bottlenecks, because, at its root, it involved companies paying deference to those with an expanded knowledge base.
Eventually, self-service entered the picture. Companies could extract intelligence without relying on other departments, and they could do so on a timetable that worked for them. Gartner’s 2016 Magic Quadrant for business intelligence and analytic platforms noted the shift toward business-user-centric tools had reached “a tipping point that requires a new perspective” — signaling that the shift was fully underway. Companies had to begin thinking about their data in different ways, especially considering that folks outside of IT needed to maintain access to what might be sensitive information.
Augmented analytics came next, and with it came basic Machine Learning (ML). With this process, insights and analysis could be woven into existing workflows, which reduced the complexity for non-IT team members and reduced the time required from extracting insights to putting ideas into practice. By 2021, Gartner’s verdict was that “differentiation has shifted to how well platforms support augmented analytics,” with ML assisting in data preparation, query generation, and insight generation. This technology was powerful, but still required humans to do much of the heavy lifting.
Today, we herald the arrival of generative AI copilots and a significant step toward autonomous, self-training, and iterative technology. Copilots also represent a major change in the way users interact with AI — questions can be asked in plain language, eliminating the need for even a rudimentary knowledge of coding. Agents take the idea further. They follow up, chase down anomalies, and trigger a response when something requires attention.
What do agentic agents change?
AI agents handle every step of the process. They carry a task from question to action, investigating details in the process to improve understanding and ensure decisions are grounded in sound data. They include humans in the process by having a specialist look at each step for modeling, data preparation, or visualization purposes.
Adoption forecasts for agentic agents are steep. Gartner predicts that by 2028, at least 15% of day-to-day work decisions will be made autonomously through agentic AI, up from zero percent in 2024. Additionally, the same Gartner research projects that 33% of enterprise software applications will include agentic AI by 2028 — again, an astounding leap up from less than one percent in 2024.
What does proper business context look like?
Agents might have the freedom to run rampant within a company’s data supply, but they still won’t know the lingo an organization uses to describe its entities. What’s an ‘active customer’, and how does it differ from an ‘archived’ one? When are workflows followed closely, and when do contingency plans come into play?
Two distinct layers fill this gap. A company’s ontology models the core business units of a company and demonstrates how they interact with one another. Think of it as an actionable organizational chart that offers clarity and context to the AI directly. Once an agent understands how things are organized behind the scenes, a semantic layer fills in the vocabulary — defining the metrics, synonyms, and instructions an AI applies throughout. Together, the ontology and semantic capabilities create a glossary that AI agents can use to understand a company’s operations and keep them on task.
Analysts and vendors are thinking along similar lines. Gartner predicts that by 2030, semantic layers will be considered critical infrastructure alongside data platforms and cybersecurity. Consider that in September 2025, Snowflake, Salesforce, dbt Labs, RelationalAI, and BlackRock launched the Open Semantic Interchange, a vendor-neutral specification for data definitions. A common semantic standard allows for AI agents to scale and translate effortlessly, increasing their shelf life and ensuring they continue to adapt.
Where is the industry heading?
As we move into 2027, we can expect that analytics will no longer feel like an outside function — instead, it will sit comfortably within tech stacks, camouflaged into the tools already used by employees. For example, a Finance lead may ask a question in a chat window, then an agent checks governed definitions and runs the analysis, and a workflow opens for approval. All of this would be visible to monitoring humans, who can track the process and iterate on changes.
Additionally, specialists will remain important, though they would work alongside agentic AI rather than as its diligent gatekeeper. Data scientists produce custom models and tailor reports, while agents handle routine investigations that used to eat up countless hours of time.
What should leaders watch out for?
- Understand that value arrives slowly. In its report The GenAI Divide – State of AI in Business 2025, MIT’s NANDA initiative found that roughly five percent of AI pilots reach rapid revenue acceleration among the organizations studied. The report identifies the causes of this general slowdown as a learning gap and flawed enterprise integration. Allow agentic AI to prove itself.
- Understand that projects get canceled. Gartner expects over 40% of agentic AI projects to fall by the wayside by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls. The research firm also flags agent washing, the rebranding of existing products as agentic, and estimates that only about 130 of the thousands of agentic AI vendors are the genuine article. With the proper time and investment, agentic AI can succeed.
- Governance lags deployment times. Gartner projects that by 2030, 50% of AI agent deployment failures will stem from inadequate runtime enforcement on AI governance platforms. To assist in curbing this trend, companies can leverage permissions, audit trails, usage and cost tracking, and a sandbox for road testing before production begins in earnest.
- Unverified output has a price. In 2025, Deloitte agreed to repay the final installment of a massive contract with Australia’s Department of Employment and Workplace Relations after it was discovered that its report contained fabricated citations. Agents that are automatically triggered by data could introduce similar errors into live transactions. All output must be verified to some extent.
What should companies do now?
Start by settling who owns business context, because the answer is still contested. Data engineers know how to build and run a semantic layer. Business subject matter experts know what ‘customer’, ‘revenue’, and ‘active’ mean inside the company, and they rarely write code.
dbt Labs’ Joey Gault writes that the people who best understand these concepts “are typically not data engineers,” while concluding that the question of who should own the layer “doesn’t have a universal answer.”
That tension is playing out in enterprise accounts, and platform roadmaps spend far more time on the agents than on who maintains the definitions beneath them. A workable split gives data engineers the plumbing, including deployment, credentials, integrations, and performance, and offers business experts the meaning of each metric.
Finance and Sales show why the split matters. Sales counts a booking at the moment a contract is signed, while Finance recognizes revenue after satisfying any outstanding obligations. Both definitions serve a purpose, and an agent asked for last quarter’s revenue will pick one unless the semantic layer says which.
To reconcile the two, give each definition its own name, such as booked revenue and recognized revenue, and document its owner, filters, and grain. Decide which definition answers an unqualified question about revenue, and assign that call to a named tiebreaker, typically the Finance organization. Record the reasoning so the next dispute starts from precedent.
The semantic layer also needs governance of its own. Every metric should carry a version, an owner, and a change history. Changes should travel a defined path, and downstream agents and dashboards should receive notice before a definition shifts. Logging which agents ran, who used them, what data they touched, and what each team spent lets a wrong answer trace back to the definition behind it.
The stakes are rising. Gartner expects half of organizations to use autonomous AI agents to convert governance policies into machine-verifiable data contracts by 2030, and those contracts depend on policies that already have owners. Gartner’s Rita Sallam advises experimenting with governance agents in low-risk pipelines before widening scope. The same discipline applies here — pick the handful of metrics agents will query most, assign owners, and prove the process before adding more.
Companies that demonstrate mindfulness today will be primed for when agentic AI matures. The decisions they make as an organization, powered by technology, will continue carrying them to whatever development happens next.
