AI adoption and capability have moved extraordinarily quickly. According to Stanford University’s AI Index Report 2026, generative AI reached 53% population adoption in just three years, faster than personal computers or the internet. More than half of employees globally (58%) reported using AI at work semiregularly in 2025, while global corporate AI investment more than doubled across the same year. Organizations are not introducing a static technology; they are harnessing a capability that is evolving faster than their normal processes for workforce transformation, governance, training, and organizational change.
At RELI, we have previously argued that organizations should start with the mission, not the model. Doing that requires understanding more than the mission objective; it requires understanding how people actually accomplish it. So how do organizations ensure that what they are implementing actually helps people do the work they need to do?
For a deeper discussion on why organizational context is essential to successful AI adoption, read our interview with Angela Mercado and Jason Balser, Why Context Matters: Turning AI Into a More Valuable Tool.
We have to talk to the people doing the work. We have to understand what they are actually doing day to day, including the things that aren’t written in the standard operating procedures (SOPs). We also need to understand whether those SOPs and the accompanying documentation accurately describe how the work is being performed.
Why wouldn’t they? The simplest answer is that work evolves. People make adjustments to increase efficiency. A step in a process changes verbally or because a new tool is introduced, and no one revisits the SOP. Often, SOPs provide step-by-step instructions without explaining why someone is doing something or why a particular step is necessary. There may be multiple documents containing contradictory guidance, leaving employees to determine how to get the work done rather than how to align with a conflicting process.
Organizations are swimming in artifacts: policies, dashboards, meeting minutes, emails, reports, Jira tickets, Teams conversations, SharePoint repositories. But all of that documentation only captures pieces of what happened.
AI gives us an unprecedented ability to begin connecting those disparate pieces. But what about the context that was never captured in those documents? After a year on a project, we may have 52 weekly meeting minutes, 300 SOPs, multiple reports, a year’s worth of Teams chats, and thousands of emails. What happens after three years? Five?
Models are increasingly capable of handling larger context windows. In fact, context windows have grown by almost 30 times per year since mid-2023. Models that once handled only a few thousand tokens can now process one million or more in a single input, equivalent to multiple books or an entire codebase in a single pass. But larger context windows do not necessarily produce deeper understanding. There remains a wide gap between the amount of context a model can accept and the amount it can meaningfully use.
If scaling the amount of information isn’t the answer, how do we identify and harness the valuable context?
Again, we have to look to the humans executing the work.
The people within an organization accumulate, interpret, and share the undocumented pieces. Organizations have long relied on employees to pass along the “unwritten rules”:
- The Deputy Administrator does not like to see stoplight colors in executive presentations because they oversimplify risk.
- A particular step was added to a process because Jan in Accounting is the person who actually executes a task, and sending it directly to her speeds things up.
- When the customer asks about a missed milestone, the team knows the concern isn’t really about this week’s delay, it is compounded by a similar issue eight months ago.
Those details may never appear in an SOP. They may not be captured in meeting minutes or reflected in a dashboard. Yet they influence how people interpret information, make decisions, and accomplish the work.
If that rationale isn’t captured, an AI system can have perfect access to an organization’s paper trail and still construct an incomplete understanding of how the organization works.
Context is the connective tissue between what an organization has documented and what its people understand. And much of that context still lives with the people doing the work.
We thought our repositories contained our organizational knowledge. AI is revealing that they contain our organizational artifacts.
