Context: The Key to Artificial Intelligence in Cultural Organizations

26 January 2026. Published by Benoît Labourdette.

6 min

Artificial intelligence can do nothing without context, a fact we are beginning to realize. The quality of an AI’s responses depends entirely on the quality of the context it is provided. For cultural organizations, this reality opens a major strategic perspective. The true challenge is not to adopt this or that AI tool, but to build the organizational and documentary conditions that will allow these tools to truly serve their missions.

The Revelation of “Context Engineering”

As we begin 2026, one observation stands out in the field of artificial intelligence. Every startup, every research lab, and every major tech company is working on the same problem. This problem has a name: context engineering.

As one Silicon Valley investor puts it: “Everyone, especially in B2B, is trying to solve the context graphing problem. AI agents need data to answer questions that arise in your world—the corporate world—and data is messy. There are emails, scattered documents, unassociated and uncleaned data everywhere.”

This field observation confirms what researchers at Anthropic have formulated. If prompt engineering was about asking better questions, context engineering is about giving the AI the foundations to answer them correctly. The distinction is crucial. You can have the best AI model in the world and ask it the most relevant question possible, but if you don’t provide the adequate context, its response will be generic, disconnected, and unusable.

Why Context is Fundamental

To understand the importance of context, one must return to what generative artificial intelligences actually are. These systems function through statistical prediction, generating responses by calculating the most probable next sequence of “tokens” (units of text). The context window is the set of information the AI can “see” at the moment it generates its response.

Recent research shows that AI models perform much better when they have access to rich context. The technical documentation for Claude (Anthropic) states it clearly: “Generally speaking, AI models tend to perform better on all tasks when they have more context.” It is not simply about providing more information. It is about providing the right information, well-structured, at the right time.

This is because of a phenomenon researchers call “context pollution.” When an AI is drowned in irrelevant information, its precision drops. Studies also show a problem known as “lost in the middle,” where models remember information located at the beginning and the end of their context better than in the middle. In other words, the quality of the context matters as much as, if not more than, its quantity.

What This Changes for Companies

This technical reality has profound implications for organizations. As an InfoWorld article notes: “The true competitive advantage for companies won’t come from picking the right model, but from building a proprietary context”—meaning the data, documents, workflows, business knowledge, and internal policies that shape the AI’s reasoning.

For agentic AI systems—those AIs that perform tasks autonomously—the need is even more critical. These agents need “a deep understanding of where they are, what they know, and the constraints that apply.” Without rich context, an AI agent can only produce generic actions that are potentially ill-suited, or even dangerous, for the organization.

Companies rich in data but poor in structured context will not be able to deploy AI beyond experimentation. This is the diagnosis made by many analysts for 2026. Organizations that fail to develop this capacity will remain stuck at the stage of gadgets and marginal productivity gains, never accessing the true transformative power these technologies can offer.

The Specific Challenge for Cultural Organizations

For cultural structures, this challenge takes on a particular dimension. These organizations produce a considerable wealth of documentation, consisting of programming, activity reports, audience feedback, project archives, correspondence with artists, mediation records, meeting minutes, evaluations, surveys... But this wealth is often scattered, not preserved, poorly filed, insufficiently structured, or not collected at all.

As I have developed in my previous work, for the enriching power of AI to exist, we humans must adopt a new way of doing things: documenting all of our organization’s activities much more thoroughly. This involves taking photos, writing narratives, recording audio, collecting documents, and properly indexing these corpora so they can be processed by artificial intelligences.

This documentation work is not a simple technical matter. It is a profound cultural shift. It requires scanning documents, naming files correctly, filing them properly, making audio recordings, applying high-quality voice recognition, establishing summaries and narratives, archiving important email exchanges, and keeping traces and versions of documents. Ultimately, it is the work of a documentalist, which is a new skill to be cultivated by humans across all professions.

Cooperation as a Prerequisite

But here is the essential point, often neglected by purely technological approaches. The creation of high-quality context is inseparable from a culture of cooperation within the organization.

Useful context for AI is not just made of existing documents. It must be actively constructed, which implies that people agree to document their practices, including their trials and errors; that they share their tacit knowledge, which is often unwritten; that they collaborate to structure information together; and that they agree on naming, filing, and summarizing conventions.

Yet, in many organizations, even the smallest ones, these cooperative practices do not exist or remain embryonic. Silos between departments, a culture of secrecy, fear of judgment, and a lack of time dedicated to documentation all stand in the way of creating the context the AI needs.

As the investor’s observation highlights: “Context problems are very tricky to solve. They exist differently in every world. Every client you visit is unlike any other. It’s a very important product problem: how do you build a product when the data is so different?”

The answer cannot be solely technological. It requires human, collective work to pool and structure knowledge.

Collective Intelligence Before Artificial Intelligence

This is why I argue that collective intelligence must precede and accompany artificial intelligence. What needs to be cultivated and developed is, first and foremost, the bond. Creating links between people, fostering cooperation, and putting collective intelligence in place—including to better use artificial intelligence, and to do things together that machines cannot do.

As Laurent Alexandre and Olivier Babeau write: “What is dying is not human work. It is standardized execution, the rigid model of lifelong employment. What is being born is an augmented, mobile, hybrid labor, supported by powerful tools but more dependent than ever on the quality of the minds that mobilize them.”

Manufacturing good context for AI involves decentralization and communication among all stakeholders in the organization. Structures that operate flexibly and agiley must reorganize into micro-hubs that cooperate and enrich one another. It is this collective intelligence that will produce the context artificial intelligence needs to be truly useful.

Three Paths for Cultural Structures

Faced with these stakes, I suggest that cultural organizations engage in an approach that coordinates three inseparable dimensions.

1. Cultivating Dynamics of Cooperation and Collective Intelligence

The first step is to establish a culture of cooperation, sharing, and documentation within the organization. This involves participatory workshops to map the skills, desires, and needs of each person regarding AI; implementing practices for collective activity documentation; identifying “AI mentors” within teams; and establishing regular rituals for sharing and capitalizing on learning. Collective intelligence methodologies—World Café, Open Space Technology, co-construction workshops—offer proven frameworks to kickstart these dynamics.

2. Making Strategic and Ethical Choices for AI Tools

The second dimension concerns the choice of the tools themselves. In the context of a cultural structure funded by public money, this choice must integrate Corporate Social Responsibility (CSR). The location of hosting, the control of these technologies, the biases they carry, and the energy they consume are all parameters to consider. Various solutions exist, from proprietary models to open-source models, and from sovereign hosting (such as Infomaniak) to local installations (using LM Studio, for example). Each organization can define a usage policy consistent with its values.

3. Implementing Procedures for Context Creation

The third dimension is the most technical. It involves concretely setting up the procedures, tools, and workflows that will allow for the creation and maintenance of high-quality context for AIs. This includes defining conventions for naming and filing documents; setting up progressive synthesis procedures (voice recognition, then synthesis, then integration into the corpus); structuring data so it is usable by AI agents; creating “context layers” accessible via APIs or standardized protocols; and training teams in these new practices.

We Are the Context

The great lesson of this technological moment is that AI has no value in itself. Its value depends entirely on what we bring to it. We are the context: our history, our projects, our values, our audiences, our expertise, our relationships. The richer, more structured, and more accessible this context is, the more useful AI can be to us.

But building this context is not a simple technical project. It is an organizational transformation that affects work practices, relationships between people, and the very culture of the institution. That is why work on the human element and work on the technical element are inseparable.

As I wrote recently: “The institutions that will survive and thrive are not those with the best technical tools, but those that have known how to maintain, enrich, and deepen the links between the people who compose them.” Technology is only an amplifier; it amplifies our capabilities as much as our dysfunctions.