But as AI gains more autonomy, an important question emerges: What happens to human agency when machines can make and execute more decisions?

This question is at the center of the keynote “As AI Gains Agency, What Happens to Ours?” from Agentic Web Week Munich 2026.

Featuring Manfred Steyer (ANGULARarchitects), Stefan Priebsch (thePHP), and Lianne Potter (NorthStart Intelligence), with journalist Frank Puscher as moderator, the session explores a fundamental challenge of the AI era: deciding where machines should have agency, where human judgment must remain essential, and what increasingly autonomous systems mean for people.

From AI Assistance to AI Agency

For years, much of the conversation around AI in software development has focused on productivity.

AI tools can help generate code, explain technical concepts, summarize documentation, and accelerate repetitive tasks. In these scenarios, the developer generally initiates a request, evaluates the result, and decides what happens next.

Agentic AI introduces a different model of interaction.

Instead of responding only to individual prompts, an AI agent may carry out a sequence of actions toward a defined goal. Depending on its configuration and permissions, it may interact with external systems and execute parts of a workflow without requiring human input at every step.

This changes the relationship between humans and AI.

The question is no longer simply how AI can support our work. It is also how much authority we are prepared to give systems that can act on our behalf.

What Does Human Agency Mean in the Age of AI?

Human agency is the ability to make choices, set goals, and influence outcomes through our own decisions and actions.

As AI systems become more autonomous, the distinction between human intention and machine execution becomes increasingly important.

An AI system may be capable of performing a task, but that does not automatically mean it should decide when or how the task is performed.

For software developers and organizations, preserving human agency means retaining meaningful control over the goals, boundaries, and consequences of automated processes.

The challenge is to benefit from AI capabilities without allowing convenience or automation to replace human judgment by default.

Where Should AI Autonomy End and Human Judgment Begin?

Not every decision carries the same level of risk.

Some activities may be suitable for a high degree of automation, while others require contextual understanding, professional responsibility, or explicit human approval.

This creates an important distinction between delegating execution and delegating decision-making authority.

An AI agent might be authorized to perform specific operations within a workflow. However, the boundaries of that authorization still need to be defined by the people responsible for the system.

For engineering teams, relevant questions include:

• Which tasks can AI systems perform independently?
• Which actions require explicit human approval?
• How can users understand what an AI agent is doing?
• What happens when an automated decision is incorrect?
• Who remains accountable for the consequences?

These questions are not simply about the technical capabilities of an AI model. They concern the design of the overall human–AI system.

Why Does Human Accountability Matter in Agentic AI?

As AI systems gain the ability to act, questions of responsibility become harder to ignore.

Traditional software workflows generally provide identifiable relationships between human decisions, system behavior, and operational outcomes.

Agentic systems can make these relationships more complex by introducing automated decisions and multi-step actions between the original human instruction and the final result.

That makes accountability an essential part of agentic AI design.

Organizations need to determine who authorizes an AI system’s actions, who can intervene when problems occur, and who is responsible for reviewing outcomes.

Delegating a task to an AI agent does not, by itself, transfer organizational accountability to the AI system.

For developers and technical leaders, this distinction matters when designing applications that interact with business processes, external services, or other systems.

How Can Humans and AI Systems Work Together?

The future of agentic AI does not have to be framed as a choice between complete automation and complete human control.

A more useful approach is to consider how responsibility can be distributed across a workflow.

AI systems can handle appropriately delegated activities, while humans retain the ability to define objectives, evaluate outcomes, and intervene when necessary.

This relationship requires clearly defined boundaries.

In practical software architecture, those boundaries may be represented through permissions, approval mechanisms, visibility into system actions, and controlled execution paths.

The goal is not merely to make AI agents more capable. It is to make their behavior understandable and manageable within the systems in which they operate.

Why Is Trust Important When AI Systems Can Act Independently?

Trust in an AI system involves more than whether it can produce a correct answer.

When an AI agent can perform actions, its reliability also depends on how it operates, which permissions it has, and whether people can understand or challenge its behavior.

This places greater importance on transparency, system boundaries, and effective oversight.

A capable agent that operates without clear limits may introduce risks even when its individual actions appear technically valid.

For users and engineering teams, trust depends on more than the promise of automation. It also depends on the ability to recognize when human intervention is needed.

What Does Agentic AI Mean for Developers and Software Architects?

Agentic AI introduces new considerations for application architecture and software development.

Developers are no longer designing only for interactions between users, interfaces, and backend services. They may also need to account for AI agents that initiate actions within those systems.

This raises questions about permissions, observability, reliability, and control.

For example, when an agent can interact with an external API, the surrounding application must determine which operations the agent may perform and under what conditions.

Similarly, AI-driven workflows need mechanisms for understanding execution outcomes and responding to unexpected behavior.

The engineering challenge is therefore not limited to integrating an AI model into an application.

It also involves designing the infrastructure and interaction patterns that allow humans and AI systems to operate together responsibly.

How Can We Preserve Human Agency as AI Evolves?

The central challenge of agentic AI is not simply increasing the number of tasks machines can perform.

It is deciding how autonomous capabilities should fit into human activities, professional responsibilities, and organizational decision-making.

As AI becomes more capable of acting and interacting in ways that resemble human behavior, clear boundaries between machine autonomy and human authority become increasingly important.

Preserving human agency does not necessarily require rejecting automation.

It requires deliberate choices about which responsibilities we delegate, which decisions we retain, and how we remain able to influence outcomes.

Conclusion: AI Is Gaining Agency. Human Responsibility Still Matters.

Agentic AI represents an important shift in how people interact with intelligent systems.

The move from AI assistance toward autonomous action creates opportunities for new applications and more flexible workflows. At the same time, it raises fundamental questions about control, trust, and responsibility.

For developers, architects, and technology leaders, the challenge is to build systems that use AI capabilities without losing sight of the people affected by their actions.

The future of AI is not defined solely by what machines become capable of doing.

It is also shaped by what humans decide to delegate—and what they choose to keep under their control.

Watch the full session below: