The organizational shift from a human institution to a computational system places data at the centre. In a computational system, data is not a reference material. It is the material. Yet for decades, data has sat behind the business rather than inside it. It served as a retrospective mirror that showed what had already happened, leaving humans to interpret it and decide what to do next. This was the practical reality of “becoming data driven,” and it depended entirely on human motivation, human time, and human interpretation.
AI agents change that relationship completely.
When work is carried out by autonomous systems, data stops being an optional. Agents do not choose whether to use data. Agents operate on it, reason through it, and depend on its quality to act correctly. They require data at the point of action, which turns data into a transaction medium: the channel through which work is initiated, validated and completed.
In this world, data is no longer a dashboard you consult. It becomes the substrate that fuels how your agentic workforce behaves, what it learns and what it can do. This is the shift from using data to understand the business to using data to conduct the business.
All of this leads to a simple conclusion: if agents depend on data to act, then organizations must build an environment where that data is accessible, interpretable, and enforceable at the moment of action. That need becomes the architectural mandate for the next evolution of data infrastructure.
[CTA_MODULE]
The future of data infrastructure
A structural shift is underway. As workflows transition from human execution to agent execution, the entire data and application ecosystem will be reshaped. This shift has 4 major components, each reshaping a different layer of the stack.
Natural language interfaces will open up data access to anyone who can express a question or instruction in everyday language. No specialist skills. No learning curve. No filing of tickets with a data team. This is the final barrier to data accessibility that the industry has struggled to overcome.
Agents will become the universal mechanism for task execution, replacing bespoke operational workflows that depend on specific systems, roles, or technical skills. Instead of tools dictating how work must be done, agents will carry work across systems and contexts. They become the unit of execution for the organization’s operations.
Operational systems and business applications will no longer be the primary place where humans carry out work or where all business logic lives. Their role shifts to enforcing the domain-specific constraints that ensure operations run correctly. They become durable, rules-oriented environments rather than user interfaces, governing definitions like what counts as a valid record, who may modify it and what must be true before any system or agent is allowed to act.
The data layer becomes the operating layer for agentic work, shifting from a passive record of what happened to the environment in which decisions and workflows occur. Current data stacks were built for humans running queries, not agents orchestrating continuous, automated workflows. Humans can negotiate ambiguity. For example, if finance and sales disagree on ARR (annual recurring revenue), they can talk it through and decide which number to use for the board deck. Agents cannot. They require explicit rules, defined precedence, and conflict resolution inside the data itself.
To support agentic operations, the data layer must evolve in 4 critical ways:
- Universal accessibility. Data must be interoperable across every system in the organization.
- A single canonical answer for every core concept. It is no secret that data leaders have avoided this job for decades. It is fiddly, political, and rarely celebrated. But deciding the right answer and where it should live is no longer optional. As Jamin Ball put it, “The companies that win this cycle will be the ones that build amazing agentic experiences on top of boring, rock-solid sources of truth.”
- Contracts for how data is allowed to change. Clear boundaries ensure agents act predictably and workflows remain stable.
- Explicit definition of meaning and precedence. The data must carry the organization’s definitions, logic, and IP, so agents can interpret and act on it without human mediation.
The data architecture that will power the next era of work
Fivetran spent more than a decade making analytics-ready data dependable and accessible. The shift to AI introduces an additional set of requirements. AI systems need more than well-modeled tables. They rely on data with traceable provenance, standardized semantics, guaranteed interoperability, and reliable availability at the point of execution.
Across our work with enterprises, we see the same shift unfolding, and several capabilities are becoming essential.
1. High quality data, complete with its lineage and history
AI begins with trustworthy, reliable, and well-structured data. In an AI-driven environment, that foundation must be paired with an understanding of where the data originated, how it was transformed and how it has changed over time. In a future state of Fivetran and dbt, organizations will gain end-to-end visibility into the origins, transformations, and lifecycle of their data. This gives AI systems the ability to assess reliability, understand provenance, and reason about whether data is fit for a given task.
2. Transformation that adds context
Meaning does not arise from data on its own. It is created through the models that express how a company operates. Data is transformed, joined, and enriched so it reflects the way the business actually works. This work encodes the organization’s intellectual property, the tribal knowledge that sits in employees’ heads but must be explicitly represented for AI to use.
3. A universal data layer that any system can consume
We are past the point where BI tools are the only consumers of data. AI agents, operational systems, and emerging tools all require direct access. Because we cannot predict every future tool or model, the most resilient approach is to store and manage data in a universal format. A universal layer future-proofs the enterprise and ensures that any new system can plug in without custom engineering work.
4. Operational access at the point of action
AI becomes valuable only when it can participate in the workflows where outcomes are determined. If AI supports sales, finance, or support, it must be able to deliver trustworthy, enriched data directly inside the applications those teams depend on. This is what moves AI from advisory to operational, from suggesting actions to taking them.
The agentic future is an infrastructure mandate
AI will not deliver competitive advantage because of the models it deploys, but because of the data foundations those models rely on. The winners will be the companies that build stable, interpretable, and interoperable foundations long before the agentic wave crests.
The future workforce is already forming. The question is whether your data is ready to support it.
For organizations preparing for this shift, we are ready to support the data foundations that enable it.
[CTA_MODULE]
