Onboarding new AI hires calls for context engineering – here’s your 3-step action plan

Follow ZDNET: Add us as a preferred source on Google.

ZDNET’s key takeaways

• Successful adoption of AI agents requires context engineering. 
• Context engineering requires access to data, metadata, process flow, and more.
• Context engineering ensures your data is ready for agentic AI usage. 

Why is it that your existing employees initially outperform the new rockstar you’ve just hired?  And why do you have a period of onboarding before a new hire gets up to speed?

Institutional knowledge. The new rockstar knows how to do the job. That’s why you hired them.  But they need time to understand the company culture, processes, approaches, applications, their team, and customers and partners. 

Also: Is that a scam? This new ChatGPT trick scans suspicious numbers, emails, and links – for free

In the AI world, the institutional knowledge is called context.  AI agents are the new rockstar employees. You can onboard them in minutes, not months. And the more context that you can provide them with, the better they can perform. 

Now, when you hear reports that AI agents perform better when they have accurate data, think more broadly than customer data. The data that AI needs to do the job effectively also includes the data that describes the institutional knowledge: context.

We’ll get to context engineering in a minute.

Understanding context

So let’s look at the different types of context, its source, and whether it’s structured or unstructured — all of which will determine how it is presented to the AI agent. 

Also: More workers are using AI than ever – they’re also trusting it less: Inside the frustration gap

You keep hearing about models having a large context window. Claude has a 1-million-token context window; ChatGPT 5.2 has a 400,000-token window. But this is not sufficient to handle everything about the company. Consider Salesforce’s org configuration — 20 apex classes of relatively high complexity are over 250,000 tokens. So, we need to be selective and provide the context for the role that the AI agent is delivering: context engineering.

Context engineering

As you can see from the table below, much of this information is unstructured. Your employees are good at interpreting it and filling in the gaps using their judgment and applying institutional knowledge. AI agents can now parse unstructured data, but are not as good at applying judgment when there are conflicts, nuances, ambiguity, or omissions. This is why we get hallucinations. 

So the context you provide needs to be complete and AI-readable. But the context also needs to be specific to the role of the AI agent, so the context window is not overwhelmed. The way to do this is to consider the end-to-end process that the AI agent is performing and use that to scope the context. That requires parsing the various applications that store the context to pull the right level of information. If we look at Salesforce’s acquisitions, it starts to make sense: Data360, Informatica, MuleSoft, and Tableau are all different forms of context at scale.

Also: Is your AI agent up to the task? 3 ways to determine when to delegate

Context in context

As we’ve said, providing the correct context to the AI Agent at the right level of detail means parsing these data sources with a clear understanding of what the end-to-end process it is trying to perform.  

This is a combination of the documented business process and the application configuration encoded in the metadata and dependencies. And this is not just about whether metadata uses other metadata, but why and how. 

The process maps provide visibility into manual activities between applications or within applications. The accuracy and completeness of the documented process diagrams vary wildly. Front-office processes are generally very poor. Back-office processes in regulated industries are typically very good. And to exploit the power of AI agents, organizations need to streamline them and optimize their business processes. This has sparked a process reengineering revolution that mirrors the one in the 1990s. This time around, the level of detail required by AI agents is higher than for humans.

Also: Gen AI boosts productivity, but only for certain developers – here’s why

The understanding of the app configuration through the metadata and dependencies is available, but it is often confused by high levels of technical debt. And it requires sophisticated analysis to be complete and trustworthy. AI agents are not yet capable of taking all the metadata and making sense of it. There is simply too much data. The only approach is to use very clever, agentic workflows of chained surgical tasks to run the analysis. 

Is your content ready for AI?

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

1. Does the information exist, who owns it, and what incentive do they have to support the project?
2. Is it up to date, and what is the process for maintaining and governing it?
3. Is it written for AI, and what changes need to be made to prevent ambiguity and confusion?
4. Where should it be stored so AI can access it, and what security and access controls should be applied?
5. How should it be structured and tagged for curation, balancing details with token usage?

Let’s look at three content types — culture, business process, and applications — and consider each in turn.

Company culture

This is the information that is typically provided to new employees during onboarding, but it is also the intangible knowledge that is absorbed over time. AI agents need all of it all at once. 

Also: Forget the chief AI officer – why your business needs this ‘magician’

• Existence and ownership: This refers to the onboarding content the organization uses, including company policies. AI agents 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. There is work they need to put in to support the project, but what are their priorities and incentives? Ideally, marketing should take the lead.
• Valid and valuable: Of all the company documentation, this is probably kept reasonably current, unless there has been a recent re-brand. If so, you need to be careful what to include. It may be up to date, but is it still relevant and valuable?
• Written for AI: The onboarding material may have been written for presenting to new starters, not for 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 that the AI agent will not pick up during onboarding and through tribal knowledge.
• Access and security: 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 searched. However, we also need to consider the security and access controls. Is there IP,  sensitive data, or personal information that should not be exposed? The security level of two or three sets of data when held in isolation may be far lower than when they are combined. Once the data is aggregated, the security level may be many levels higher based on the insights that are now available.
• Structure and tagging: 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. There is a balance between providing all the information at a detailed level and the cost and feasibility of the token usage. So you need to consider how to categorize the data so that it can be sliced and diced and served up to the AI agent in the most token-efficient manner.

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 and on which it relies or delegates. 

Also: Stop using ChatGPT for everything: My go-to AI models for research, coding, and more (and which I avoid)

• Existence and ownership: Most organizations have processes documented. In 30+ years of working in business process engineering, we’ve found that processes are normally incomplete, out of date, and in a variety of formats. Fortunately, you do not need to get every process up to date; only the ones related to the AI agent you are building, which are likely 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 handle nuances, gaps, and ambiguity as well as humans do. 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.
• Valid and valuable: 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. 
• Written for AI: AI is very good at understanding process-related diagrams and procedural documents. The issue is the quality of the documentation: its completeness, accuracy, and currency.
• Access and security: Again, if it’s unstructured documentation, such as images, a solution like Data 360 is needed to make it accessible and easily searchable. But process diagrams could be presented as structured JSON, which is more easily consumed by AI.
• Structure and tagging: This is very specific to the scope and outcome of the AI agent. Therefore, 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 that describe how applications work together. Also included in these diagrams could be how the agents work together. 

Also: 5 ways you can stop testing AI and start scaling it responsibly in 2026

• Existence and ownership: This data is stored within every application as metadata. However, it needs to be more than a list of metadata. It needs to include metadata dependencies, such as the metadata analysis that Elements.cloud produces for Salesforce. An application like Informatica is designed to store metadata from multiple systems.
• Valid and valuable: The metadata is 100% accurate. The metadata analysis can be performed whenever it changes, so it can also be 100% accurate.
• Written for AI: Metadata is highly structured, and therefore, ideally suited for being read by AI.
• Access and security: 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.
• Structure and tagging: 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.

Only 7% of communication is words 

There is the common expression that communication is only 7% words. So what about the other 93%?

• The words (7%): Verbal content (the literal meaning).
• The tone (38%): Voice quality, pitch, and volume.
• The visuals (55%): Facial expressions and body language.

Tonality acts as the punctuation of spoken language. Let’s take the simple sentence: “I want to see you in my office.” Context is the 93%. We instruct AI with words; the 7%. Is it any wonder we get hallucinations and inconsistent results? We need to provide the other 93%. The context. This could include: the relationship between the customer and company, the relative importance of different aspects of the data, the stage in the process, the urgency, and the value of the outcome. And that context is provided as words and data. So we need to make sure that there is context for the context.

Also: The best AI chatbots of 2026: I tested ChatGPT, Copilot, and others to find the top tools now

Context engineering is a new term for AI agents, but the content already exists within organizations as institutional knowledge that people absorb over time. AI agents are built to accept a firehose of information but require it to be accurate and unambiguous. That has implications for organizations that want to tap into the benefits of AI agents capable of delivering sophisticated outcomes. Here is a 3-step action plan:

1. Document the scope of your AI agents, including the end-to-end process and outcomes.
2. Identify the critical contextual information required for AI agents to perform at the highest levels, and review their quality.
3. Format the contextual information in the platforms that can curate it for AI agents.

This article was co-authored by Ian Gotts, senior research fellow at Keenan Vision, co-founder of Elements.Cloud, 10X author, tech advisor, speaker, and investor.