AI & the democratization of information
Most enterprises do not have a data problem. They have an information problem.
Over the last several decades, organizations have invested heavily in systems that capture, store, and manage data. Customer interactions are recorded in CRMs. Employees are tracked in HR systems. Security events are logged by dozens of tools. Documentation lives in wikis, repositories, ticketing systems, and collaboration platforms. Every business function has acquired its own systems of record and, with them, its own growing collection of data.
The result is that enterprises now possess extraordinary amounts of data about themselves, their customers, their employees, and their operations. Yet despite this abundance, many of the most important decisions inside an organization still require significant manual effort to make.
The reason is straightforward: data and information are not the same thing.
Data consists of facts, events, records, and observations. Information emerges when those facts are organized, connected, and contextualized in a way that helps a human understand a situation and make a decision. While enterprises have become exceptionally good at collecting data, they remain relatively poor at transforming that data into information. Information is the amalgamation of data - oftentimes disparate data - and human insight.
This transformation is where much of modern knowledge work takes place.
Consider a simple question that might arise inside a security organization: should a particular employee retain access to a sensitive production environment? The answer is unlikely to exist in a single system. Relevant information may be spread across an identity provider, an HR platform, a ticketing system, a source code repository, an access management tool, and internal documentation. Someone must determine the employee’s current role, understand why access was granted, evaluate whether it is still required, assess the associated risk, and identify any relevant exceptions or approvals.
The decision itself may take only a few minutes. Gathering the information required to make the decision may take hours.
The same pattern appears throughout the enterprise. Managers preparing for planning reviews spend time collecting updates from multiple teams. Executives preparing for board meetings assemble information from finance, sales, operations, and product organizations. Security teams investigating incidents pivot across numerous systems attempting to reconstruct what happened. In each case, the act of making the decision is often less expensive than the process of assembling the information necessary to support it.
Viewed through this lens, a significant portion of what we call knowledge work is actually information assembly. Humans act as the integration layer between disconnected systems. They retrieve data, often from many tools, reconcile inconsistencies, identify missing context, and construct a coherent understanding of a situation. Much of the value they create comes not from accessing individual pieces of data, but from connecting them.
Historically, this process has been difficult because data has been fragmented across organizational and technological boundaries. Each system contains only part of the story. Understanding the whole requires navigating multiple tools, multiple teams, and multiple workflows. Over time, expertise often becomes intertwined with the ability to locate information. The people who know where data lives, which systems matter, and how seemingly unrelated pieces of context fit together become disproportionately valuable.
This is where large language models become particularly interesting.
Much of the public discussion around AI focuses on automation. Can models write code? Can they answer support tickets? Can they generate reports? These are important capabilities, but they may not represent the most significant impact of the technology within the enterprise.
A more consequential development may be the emergence of an intelligence layer that sits above systems of record. Rather than replacing existing applications, this layer interacts with them. It retrieves data from multiple sources, understands relationships between entities, synthesizes context, and presents the resulting information in a form that humans can consume.
That’s something that LLMs can do really well.
What makes LLMs particularly well suited to this role is that they are not limited to consuming information that has already been assembled for them. They can actively participate in the process of gathering it.
Traditional enterprise software tends to operate within well-defined boundaries. A dashboard queries a database. A search engine indexes documents. A reporting system aggregates metrics. Each tool is designed for a specific source of data and a specific mode of interaction.
LLMs are different because they can participate in both stages of the process: gathering information and interpreting it.
A modern model can formulate SQL queries against a data warehouse, call APIs exposed by SaaS applications, retrieve documents from knowledge bases, search source code repositories, inspect Jira tickets, review identity records in Okta, analyze logs in Splunk, and correlate information across all of these sources. To the user, this appears as a simple question. Under the hood, however, the model may be interacting with dozens of systems and executing hundreds of retrieval operations.
This matters because enterprise information rarely exists in a single place. Meaningful business questions almost always span multiple systems of record. The answer to a question such as “Who has production access they no longer need?” does not live in Okta, Jira, GitHub, AWS, or an HR system. It emerges from the relationships between them.
Historically, humans performed this work manually. They knew which systems to inspect, which reports to run, which APIs to query, and how to reconcile the results. Much of what we call institutional knowledge was, in practice, knowledge about where information lived and how to assemble it. Experienced employees became valuable not simply because they knew more, but because they knew how to navigate the organization’s fragmented information landscape.
LLMs can increasingly perform that assembly themselves. Given access to the appropriate tools and permissions, they can determine which systems are relevant to a question, retrieve the necessary data, identify relationships between entities, and synthesize the results into a coherent explanation. The same system that gathers the information can also transform it into something a human can understand and act upon.
That combination is what makes this technology significant. Enterprises already possess tools capable of querying databases, searching documents, and generating dashboards. The challenge has never been finding individual pieces of data. The challenge has been converting hundreds of disconnected observations into a coherent understanding of what is happening and why it matters.
In many ways, this is the first technology capable of operating directly on information rather than merely storing, retrieving, or visualizing data. It reduces the cost of transforming raw observations into context, context into understanding, and understanding into informed decisions.
The implications extend beyond productivity.
Information inside organizations has historically been distributed unevenly. Access often depended on tenure, role, relationships, or simply knowing where to look. Employees spent years accumulating institutional knowledge that allowed them to navigate the organization’s information landscape effectively. The ability to find information became almost as important as the ability to use it.
An LLM-powered intelligence layer changes that dynamic. When information can be assembled on demand from across the enterprise, access becomes less dependent on knowing which report to run, which dashboard to consult, or which colleague to ask. More people gain the ability to understand situations that previously required extensive investigation and organizational familiarity.
In that sense, AI has the potential to democratize information. Not because it creates new knowledge, but because it dramatically lowers the cost of accessing and synthesizing the knowledge that already exists within an organization.
Importantly, democratizing information is not the same as democratizing judgment. A model may be able to assemble relevant context, identify relationships, and summarize findings, but judgment remains a human responsibility. Organizations still need people to evaluate tradeoffs, assess risk, determine priorities, and make decisions under uncertainty.
If anything, easier access to information increases the value of judgment. As information becomes broadly available, competitive advantage shifts away from information gathering and toward decision quality. The bottleneck moves from obtaining context to interpreting it.
For decades, enterprises have focused on collecting and storing data. The next phase will be about making that data accessible, understandable, and actionable. The organizations that succeed will not necessarily be those with the most data. They will be those that can most effectively transform data into information and place that information in the hands of the people responsible for making decisions.
That is why I suspect one of the most important applications of AI inside the enterprise will not be automation. It will be reducing the cost of information itself—and in doing so, increasing the value of human judgment.