Why organizations should treat AI workers like teammates

Over the past few months, I’ve noticed more and more organizations experimenting with AI agents. While AI was initially used primarily as a tool for generating text, code, or analyses, we are now seeing a new development emerge. AI systems are not only gaining access to more and more information, but they are also gaining the ability to perform actions independently. They process documents, answer questions, analyze data, write code, launch workflows, and in some cases, even make decisions within pre-defined frameworks.

The Noma Security GitLost case

That development was recently highlighted again by Noma Security’s GitLost case. Researchers demonstrated how an AI agent could be manipulated into retrieving information from private repositories and making it publicly available. Many analyses rightly focused on prompt injection, model security, and guardrails. Yet, another observation stuck with me. GitLost doesn't just show that AI introduces vulnerabilities; it primarily shows that we are still treating AI agents like software, while in practice, they are increasingly behaving like colleagues. [noma.security]

That might sound like a semantic debate, but it has major consequences for how we design, implement, and govern AI. For years, we’ve viewed software as a tool. An application performs a pre-defined task. The logic is hardcoded, and the behavior is largely predictable. AI agents function fundamentally differently. They read information from various sources, interpret context, combine knowledge, use tooling, and perform actions based on that. Not because they have consciousness, but because we are increasingly giving them roles previously reserved for knowledge workers.  

It primarily shows that we are still treating AI agents like software, while in practice, they are increasingly behaving like colleagues.

Think of AI as a digital colleague

As soon as you view AI as a digital colleague rather than a piece of software, the discussion changes. When a new employee starts on their first day, we don't automatically give them access to all systems, documents, and company information. We invest in onboarding first. We explain how the organization works, what information is relevant, how decisions are made, and what responsibilities come with the role. No one finds it strange that a new colleague needs context before they can act independently.

With AI, we often do the exact opposite. We connect a model to data, provide access to tooling, and then expect the system to produce valuable results. In doing so, we might be skipping the most important step. After all, good decisions don't come from data alone. Good decisions come from context.

AI-First Framework as a foundation

That premise also forms the basis of the AI-First Framework that we developed within TeamValue. The core principle of that framework is that AI autonomy is only possible with complete, explicit, and reliable context. An AI worker shouldn't just have access to information; it must also understand which information is authoritative, who owns that information, what trade-offs were made previously, what decision-making frameworks apply, and within what mandate the system is allowed to act. Without that context, an AI can generate answers, but it cannot make reliable decisions.  

From that perspective, GitLost becomes much more interesting. The attack didn't succeed because the model wasn't intelligent enough. The attack succeeded because an agent could combine unreliable input with access to sensitive information and was given the authority to act independently. The vulnerability, therefore, wasn't just in the prompt, but primarily in the lack of sufficient frameworks surrounding autonomy, access, and context. What ultimately determined the impact wasn't the model itself, but the combination of permissions, scope, and room for action that the agent had been granted.  

AI requires breaking old patterns

For architects, this is actually a familiar pattern. When organizations moved to the cloud en masse, we discovered that infrastructure wouldn't scale without Landing Zones. When data became a strategic company asset, we discovered that self-service analytics wouldn't scale without governance and DataOps. With AI workers, we are seeing the same development emerge again. Not because AI is unique, but because every form of autonomy places new demands on ownership, governance, and control.

That is exactly why we at TeamValue talk about The Golden PathNot as a collection of technical requirements or architectural principles, but as a BizDevOps way of working where business, development, operations, and increasingly AI workers collaborate within the same framework of accountability, transparency, and continuous improvement. The challenge is shifting from implementing AI to organizing autonomy.

Onboarding and collaborating with digital colleagues

That might just be the most important lesson right now. The question is no longer which model we use or which prompt techniques work best. The question is how we onboard, guide, authorize, and manage our digital colleagues. How do we ensure they have the right context? How do we make decision-making explicit? How do we organize ownership? And how do we ensure that autonomy remains manageable when applied at an ever-increasing scale?

The organizations that successfully scale AI won't necessarily be the ones with the most advanced models. They will be the ones that best understand that autonomy without context leads to poor decisions, that autonomy without ownership leads to risks, and that autonomy without guardrails eventually leads to a loss of control.

Perhaps that is the next step in modern platform architecture. First, we learned to standardize infrastructure securely. Then, we learned to standardize data securely. Now, we are learning how human and digital colleagues can collaborate securely.

How is your organization preparing for AI workers as full-fledged digital colleagues?

AI Inspiration Lunch

On Thursday, October 8th from 12:00 PM to 2:00 PM (PEC Zwolle Stadium), we are hosting an AI Inspiration Lunch. During this lunch, we will guide executives, directors, and managers focused on AI, innovation, and strategy through a vision that is closer than you might think. A future where employees have their own AI assistant and independent access to the knowledge, processes, and information they need. We’ll show you the building blocks required and take you step-by-step through the AI journey: from information management and governance to the practical application of AI in daily work.

Sign up for the AI Inspiration Lunch.