What Is Context Engineering? How to Make AI Agents Enterprise-Ready

Context Engineering explained

In my previous blog, I explored the concept of Context Engineering and how it is making AI Agents more reliable and, therefore, able to perform complex tasks.

With Context Engineering curating the right level of contextual information for AI Agents based on an end-to-end process, we can see how we can make AI Agents trusted enough to deliver broader use cases in the enterprise. At the moment, AI Agents are being used in narrow use cases where the contextual decision making is limited e.g. Intercom’s Fin for Support, Salesforce SDR, and Amazon’s Product Finder called Rufus. And now Salesforce is building the software stack that can provide this trusted context that will pave the way to AI Agents delivering powerful end-to-end outcomes that span more than just Salesforce.

Salesforce’s view is that trust context is the critical data that AI Agents need to be able to deliver reliably. We agree.

So the data that everyone talks about when thinking about AI Agents is more than customer data. It is in a number of different forms. Below is a table that summarizes the different data types, or content, that need to be provided to an AI Agent to provide the context, based on the type of content and the likely source. The content may be structured – so easy for an AI Agent to consume – or unstructured  – which means it needs to be delivered via an application like Data 360 (Data Cloud) that can chunk it. 

The other issue is the context window – the number of tokens – that the LLMs are getting very large. Claude is now 1 million tokens. But this is still far too small to be able to “dump” all the organization’s content into. The application metadata is hugely rich.  For example, 20 typical Apex classes in Salesforce could be 250,000 tokens.  Therefore, the content needs to be curated based on the scope of the AI Agent, and this is best thought of as the end-to-end process that the AI Agent is trying to deliver.

Let’s look at the table below and consider the “quality” of the “data”.

Content / Data type	Source	Structured / Unstructured
Company culture	Annual reports Marketing brand guidelines New employee handbook	Unstructured
Business operations	UPN process diagrams	Unstructured
App configuration	Metadata & dependencies	Structured
Data	CRM, ERP apps	Structured
Team	Org chart Job descriptions	Unstructured

Is your content ready for AI?

For each type of content, we need to ask 5 questions

• Does the information exist and who owns it?
• Is it up to date and what is the process to maintain it?
• Is it written for AI and what changes need to be made?
• Where does it need to be stored so AI can access it?
• How does it need to be structured so it can be curated?

Let’s look at each content type in turn and consider each question.

Company Culture

This is the information that is typically provided to new employees in onboarding. It is then the intangible knowledge that is absorbed over time. AI Agents need all of it all at once. 

Does the information exist and who owns it?

This is whatever onboarding content the organization uses. It could include company policies. AIAgents don’t care how dry the content is. It can also be other documents that show the culture and personality of the organization; the marketing brand guidelines, annual reports and shareholder presentations. Even the style of customer testimonials, marketing videos and office design will provide color to a blank canvas. Maybe there is a table of corporate acronyms. The complication is that this is owned by different teams. But marketing should take the lead.

Is it up to date?

Of all the company documentation, this is probably kept reasonably current, unless there has been a recent brand change. If so, you need to be careful what to include.

Is it written for AI?

The onboarding material may have been written for presenting to new starters, not reading. Hence, there could be huge context gaps, which need to be filled with a preamble or notes. Other content needs to be set in its own context.

For example, the AI Agent needs to be told how to interpret customer testimonial videos or brand guidelines. Company policy documents are often written for humans with nuances and assumptions based on the onboarding and tribal knowledge, that the AI Agent will not pick up.

Where does it need to be stored so AI can access it?

This content is mostly unstructured and high volume. The customer testimonials may need to be transcribed to text rather than indexed as video. This means that a solution like Data 360 needs to be used to make it accessible and easily searchable.

How does it need to be structured or tagged so it can be curated?

This is difficult data to structure, as virtually all of it is required as background for the AI Agent rather than in the context of delivering a process.

Business operations / process

The documented business processes are the critical structure for the AI Agent to deliver an outcome. But they also describe the supporting processes that surround the AI Agent that it needs to rely on or delegate to. 

Does the information exist and who owns it?

Most organizations have processes documented. In 30+ years of working on business process engineering, I’ve found they are normally incomplete, out of date and in a variety of different formats. Fortunately, you do not need to get every process up to date. Only the ones that are related and tangential to the AI Agent you are building, so they are probably owned by one or two business units.

The processes need to cover both the automated and human activities, but at a far greater level of detail. AI Agents do not deal well with nuances, gaps and ambiguity the way that humans can. Now you can use AI to help you build the first cut process. It can generate process diagrams from notes, diagrams or even systems metadata. These can be refined by working with leaders and users.

Is it up to date?

I’ve always said the most important process to document and optimize is the process of process improvement. This becomes critically important for AI Agents that will take content literally, and rely on up-to-date processes documentation to behave as expected. 

Is it written for AI?

AI is very good at understanding process-related diagrams and procedural documents. The issue is the quality of the documentation: completeness, accuracy, and currency.

Where does it need to be stored so AI can access it?

Again, this is unstructured documentation; if it is in images, a solution like Data 360 needs to be used to make it accessible and easily searched. But process diagrams could be presented as structured JSON, which is more easily consumed by AI.

How does it need to be structured or tagged so it can be curated?

This is very specific to the scope and outcome of the AI Agent. So the metadata of the process diagrams is important. 

Application configuration

The application metadata describes the data structure, business logic and permissions of a specific application. If AI Agents span application boundaries, the content needs to be augmented by architectural diagrams, which describe how applications work together. Also included in these diagrams could be how the agents work together. 

Does the information exist and who owns it?

This data exists inside every application as metadata. However, it needs to be more than a list of metadata. It needs to include dependencies between metadata, like the metadata analysis that Elements.cloud produces for Salesforce. An application like Informatica is designed to store metadata from multiple systems.

Is it up to date?

The metadata is 100% accurate. The metadata analysis can be performed whenever it changes, so it can also be 100% accurate.

Is it written for AI?

Metadata is highly structured, and therefore it is ideally suited for being read by AI.

Where does it need to be stored so AI can access it?

As it is highly structured it can be stored in any database. What is critical is how it is structured so that it can be accessed. The issue is that any application has far too much metadata and it will overwhelm the token limits.

How does it need to be structured or tagged so it can be curated?

The metadata needs to be related back to the operational business processes that the AI Agent is delivering, and the data sources that the AI Agent needs.

Data

This is what is considered critical to the consistent, reliable actions of an AI Agent. But this is only one part of the puzzle. The other context we’ve referenced is the rest of the pieces. But data is the heart of it.

Does the information exist and who owns it?

The data is owned by the business units, but it may span multiple systems. The metadata and ERD will explain how to navigate the data structures to get the required data.

Is it up to date?

Data quality is an issue for every organization. Like processes, you only need fix the data that is relevant so that the AI Agent can function consistently. 

Is it written for AI?

Yes. Data is structured

Where does it need to be stored so AI can access it?

It could be accessed from the application where it is maintained. Or it may need to be aggregated by using a solution like Data 360. Certainly, AI token limits mean that not all data can be used, so it needs to be filtered.

How does it need to be structured or tagged so it can be curated?

The process that the AI Agent is navigating will help identify how to filter it or curate it

Team

The organization structure, and skills and responsibilities of team members may not seem like they are necessarily relevant. But the AI Agent is probably replicating or supporting the work of existing team members. Also, when the AI Agent cannot deliver the outcome, it will need to hand off to a human. So the Org chart matters.

Does the information exist and who owns it?

HR will keep the Org chart and employee information up to date and it will also be held in an employee or payroll system. 

Is it up to date?

Yes.

Is it written for AI?

It is structured information and used in a highly structured way, so it is ready for AI.

Where does it need to be stored so AI can access it?

The data may be in several systems, or it could all be in the HR system.

How does it need to be structured or tagged so it can be curated?

The process diagrams will have responsibilities (RASCI) against processes so it is easy to curate the information for the AI Agent.  RASCI – Responsible, Accountable, Supportive, Consulted, Informed.

Setting context

It all starts with the AI Agent input, process and outcome. That sets the context for the context. You can decide what context data is required. From there, you can then answer all 5 questions for all 5 types of context data.