Preparing your Team for the Context Layer

·Matt Barker
Preparing your Team for the Context Layer

Agents are only as capable as the systems they can reach, and the context layer is what fills the gap between your AI agents and your business applications.

In my previous posts, I explored the importance of the context layer, along with some of the challenges organisations face when giving AI access to enterprise data.

In this post, I want to focus on something more practical: how organisations can start preparing themselves to actually get value from it.

The Prerequisites: Humans, Agents and APIs

The first requirement is simple: you need people inside your organisation who are motivated to experiment with AI systems.

Start with a small group of technically curious individuals and give them access to modern agent tooling like Claude Code. Encourage them to explore real use cases, not just generic prompting exercises. The people already experimenting with AI internally will often become your best source of ideas, energy, and practical feedback.

The second requirement is APIs. Organisations with mature, well-documented APIs will move significantly faster. Ideally, APIs should have clean specifications, clear input and output schemas, up-to-date descriptions, and predictable authentication models.

The better your APIs are specced, the easier it becomes to expose them safely and effectively to AI systems.

Adopt an Executive Sponsor

Most of our early customer conversations have included an executive sponsor to help shape discussion around business priorities and AI outcomes.

No matter how interesting an individual use case may seem, if it doesn’t align with broader business strategy, it’s probably not the right place to start. An executive sponsor can help determine what the organisation is primarily focused on e.g.

AI workflows also tend to cut across teams, systems, and operational boundaries surprisingly quickly, so executive sponsorship becomes important earlier than you’d expect.

Fortunately, most executive teams and boards already understand how strategically important AI enablement is, so alignment is usually achievable.

Identify Your Use Cases

The best place to start is the people already trying to streamline their jobs with AI.

Individuals who are using tools like Claude naturally begin identifying tasks that could be done better if AI had secure access to more systems. They are often the first people who can clearly articulate where the friction exists today and what better outcomes might look like.

Another useful signal is where “Skills” are starting to appear organically inside the organisation. In practice, Skills are reusable instructions that guide agents through specific workflows or tasks.

After mentioning Skills to one CEO, he asked around internally and discovered his CMO was already experimenting with them to automate parts of the company’s marketing workflows. These bottom-up experiments are beginning to emerge across marketing, sales, finance, operations, and support teams. That’s important because it often signals genuine demand before formal AI strategy catches up.

Once you have identified potential use cases - and assuming you have executive alignment - you can begin prioritising based on business impact, feasibility, and strategic relevance.

Document Your Tasks and Decide Where to Start

Once you’ve identified candidate use-cases, the next step is to document them properly. We’ve found it useful to think in terms of the following::

This is usually the point where organisations need someone who understands both AI workflows and the underlying implementation constraints. A large part of the emerging “Forward Deployed Engineer” trend appears to be solving exactly this problem: translating business ambition into technically achievable agentic systems.

Once you have alignment between business strategy, use case, technical feasibility, and ownership, you can turn your attention to implementation.

Bonus: Codify Your Knowledge

When AI starts interacting with enterprise systems, well-documented domain knowledge becomes extremely valuable as a guide for the agent. We recommend that teams begin consolidating their operational knowledge into shared repositories.

In many cases, this doesn’t need to be sophisticated. It can simply be a well-structured Google Drive folder, an internal wiki, or a collection of operational documents.

A useful way to think about it is: “What would a new team member need in order to successfully complete this task if they joined the company tomorrow?”

This information already exists informally - in private notes, Slack conversations, and internal tribal knowledge. The proliferation of AI is simply forcing organisations to formalise and operationalise it.

Over time, this knowledge becomes the foundation for building better agent Skills and ultimately getting more reliable AI outcomes.

Conclusion

The organisations succeeding with AI today are rarely starting with massive transformation programmes. Instead they’re identifying motivated teams, documenting APIs, consolidating domain knowledge, and incrementally building the context layer required to support their agents.

The technology itself is moving extremely quickly, but many of the foundational organisational patterns are surprisingly straightforward.

In a future post, I’ll explore what the technical implementation of an enterprise context layer looks like in practice and some of the architectural patterns emerging around MCP servers, Skills, identity, and policy enforcement.