Introduction

The way we build, manage, and interact with data solutions is undergoing a fundamental transformation. For years, developers have relied on manual coding, modelling, and configuration to create powerful reporting experiences. Now, we are entering a new era: agentic development, where AI agents move beyond simply providing suggestions and begin actively contributing to the development process.

Rather than replacing developers, agentic AI changes the role of the developer. The focus shifts from manually completing every task to designing workflows, defining boundaries, and guiding intelligent agents to deliver solutions faster and more efficiently.

A useful analogy is the evolution of lawnmowers. Traditional development is like using a manual push mower — every movement and adjustment is controlled by the person operating it. Agentic development is closer to using a robotic mower. The human role changes from doing all the physical work to defining the boundaries, setting the rules, and ensuring the outcome meets expectations.

In Power BI development, this means moving from writing every measure, relationship, and configuration manually towards managing AI-powered agents that can handle increasingly complex tasks.

The Ladder of AI Control

Adopting agentic development is not an overnight transformation. Instead, organisations will progress through different levels of AI involvement, gradually increasing the responsibility given to agents.

1. Advisory Chat

At the first stage, AI acts as a consultant. Tools such as ChatGPT or Claude can help developers brainstorm ideas, troubleshoot issues, explain concepts, or suggest improvements.

The developer remains fully responsible for implementation, but AI becomes a valuable source of guidance and expertise.

2. Pair Programming

The next step is collaboration between developer and agent. Instead of simply answering questions, the AI actively supports development tasks.

Examples include:

• Renaming columns across multiple datasets
• Creating or updating relationships
• Generating DAX measures
• Reviewing and improving existing solutions

The developer remains in control but benefits from increased speed and productivity.

3. Semi-Autonomous Agents

At this stage, developers begin delegating larger pieces of work. An agent may be asked to create a complete feature, build part of a reporting solution, or implement changes independently based on defined requirements.

The developer’s role shifts towards reviewing, validating, and refining the output rather than creating everything manually.

4. Fully Autonomous Agents

The final stage is where agents take responsibility for delivering entire products or solutions.

While this represents the greatest productivity opportunity, it also introduces significant challenges around security, governance, permissions, and accountability. Organisations will need strong controls to ensure agents operate safely and effectively.

The Building Blocks of Agentic Development

For organisations looking to embrace agentic development, four key components enable agents to work effectively.

1. Context Engineering: Giving Agents the Right Knowledge

An AI agent is only as effective as the information it has access to.

Context engineering defines what an agent can see and understand. By providing information about semantic models, business rules, coding standards, and development practices, organisations give agents the knowledge they need to operate effectively.

This creates a form of working memory, allowing agents to make better decisions and produce outputs aligned with organisational standards.

2. Model Context Protocol (MCP): Expanding What Agents Can Do

If context defines what an agent knows, Model Context Protocol (MCP) defines what an agent can do.

MCP acts as the agent’s toolbox, providing structured ways for AI systems to connect with external tools and environments.

For Power BI developers, this could include connecting an agent directly to Power BI Desktop models, allowing it to analyse and modify solutions in real time rather than simply suggesting changes.

3. Agent Skills: Teaching Agents How to Work

Agent skills provide structured instructions that guide how an agent behaves.

Typically stored in files such as skill.md, these instructions define processes, standards, and best practices the agent should follow.

A key concept here is progressive disclosure. Instead of overwhelming the agent with every possible instruction, relevant information is provided only when needed. This allows agents to remain efficient while still accessing deeper technical guidance for complex tasks.

4. CLIs and Local Software: Defining the Agent’s Reach

Command-line interfaces (CLIs) and local software integrations determine which applications and systems an agent can interact with.

These capabilities extend the agent beyond conversation, allowing it to execute tasks, automate processes, and work directly within development environments.

Why PBIP Is a Game-Changer for Power BI Development

One of the biggest enablers of agentic Power BI development is the introduction of the Power BI Project (PBIP) format.

Traditionally, Power BI files were stored in binary formats, making it difficult for external tools — including AI agents — to understand and modify their contents.

PBIP changes this by representing Power BI solutions as text-based files. Instead of seeing a single packaged file, developers and AI agents can access individual components such as:

• Report definitions
• Semantic model metadata
• Visual configurations
• Relationships
• Measures

This makes Power BI development much more compatible with modern software engineering practices, including source control, automation, and AI-assisted development.

An AI agent can now identify specific changes, such as modifying a visual property or updating metadata, and make those changes directly with greater accuracy and transparency.

Governance: The Responsibility Behind Agentic Development

As AI agents become more capable, they must be treated as users within the data ecosystem.

Just like human users, agents require appropriate permissions, access controls, and governance frameworks. Organisations need to carefully consider:

• What data an agent can access
• Which actions it is allowed to perform
• How changes are reviewed and approved
• Who remains accountable for the final outcome

The future is not about allowing AI to operate without control. It is about creating a collaborative model where humans define objectives, establish boundaries, and maintain accountability while agents handle repetitive and complex execution.

Preparing for the Agentic Future

The move towards agentic development is already underway. Developers can begin exploring this future today by experimenting with tools such as GitHub Copilot, Claude Code, and Power BI-focused MCP integrations.

The journey starts with small steps: using AI for assistance, progressing towards collaboration, and eventually managing a workforce of intelligent agents.

The future Power BI developer will not simply be someone who builds reports. They will become someone who designs systems, manages AI capabilities, and guides intelligent agents to deliver better data experiences.

Agentic development is not removing the developer from the process — it is elevating their role.