1:44:21
BC TechDays 2026: Opening Keynote
mibuso.com
Overview
This keynote introduces the evolution and application of AI agents, particularly within the Business Central ecosystem. It contrasts earlier AI features like CoPilot with more advanced AI agents capable of end-to-end process management. The presentation details the technological advancements, specifically the increasing 'time horizon' of AI models, which enables more complex agent functionalities. It showcases native Business Central agents, agents built with CoPilot Studio, and engineering agents that assist in software development, highlighting practical examples and the tools available for creating and deploying these agents.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- Early AI features (2024) like CoPilot offered narrow assistance, requiring end-users to manage the entire process.
- In 2025, Business Central introduced AI agents (e.g., Sales Order Agent, Payables Agent) which take responsibility for end-to-end processes, involving users only for specific approvals or input.
- The current focus (2026) is on 'agentic AI,' with agents becoming ubiquitous across industries, including Business Central and software engineering.
Understanding this progression helps learners grasp the increasing autonomy and capability of AI tools, moving from simple assistants to sophisticated agents that can manage complex workflows.
CoPilot features in 2024 provided suggestions, but the user still performed all steps; agents in 2025 and beyond handle the entire process, like the Sales Order Agent or Payables Agent.
- The 'time horizon' of a software task measures how long a human would take to complete it, used by Meta to evaluate AI models.
- AI models have shown exponential improvement in their time horizon: GPT-3.5 (15 seconds human time), GPT-4 (53 seconds), Claude 1 (4 minutes), Opus 4.0 (20 minutes), Gemini 3.1 Pro (90 minutes).
- This increasing capability is why agents, requiring more complex AI models, were not feasible in earlier years (e.g., 2024) but are now central.
This concept provides a tangible way to understand the rapid progress in AI model capabilities and explains the technical feasibility behind the shift towards more powerful AI agents.
A task taking a human 20 minutes to complete would not be delegated to an AI with a 15-second time horizon, but would be delegated to an AI with a 90-minute time horizon, as it becomes a time-saver.
- Business Central offers native agents like the Sales Agent and Payables Agent, deeply integrated into the user experience.
- Partners can also build and deploy their own agents for Business Central, extending functionality.
- The creation process involves an agent design experience within the client and an SDK for packaging as extensions.
This demonstrates how AI agents are being embedded directly into business applications and how the ecosystem supports customization and innovation through partner-built solutions.
A partner's 'Manifest Agent' automates processing of shipping manifests from emails, extracting data to create purchase orders in Business Central, saving significant manual effort and reducing errors.
- An agent is defined by what it can see (data access), what it's allowed to do (permissions), and its instructions (guidelines and steps).
- The agent design experience offers a wizard-like interface to define these aspects, starting from templates or scratch.
- Agents are triggered by tasks and run in the background, with task logs available for monitoring and debugging.
- The process is iterative: define, test, revise instructions, permissions, and views, then package and deploy.
This provides a practical guide to how users or developers can leverage the platform to create custom AI agents tailored to specific business needs, fostering efficiency and automation.
Creating a 'Sales Return Agent' involves defining its access to customer receivables, setting permissions, and providing instructions on how to process return requests, create credit memos, and handle approvals.
- CoPilot Studio offers a flexible environment with connectors to thousands of apps, enabling agents to run autonomously or interactively.
- Business Central acts as an MCP (Model Context Protocol) server, allowing CoPilot Studio agents to interact with Business Central data.
- Configuration involves connecting to a Business Central environment, specifying the company, and defining an MCP server configuration (toolsets) that dictates exposed APIs and permissions.
This highlights how external AI platforms can integrate with Business Central, providing a broader range of tools and a more configurable agent-building experience.
A CoPilot Studio agent can be configured with a Business Central MCP connector to query customer financial details, sales orders, and invoices, and then use code execution tools to visualize the retrieved data.
- Software engineering is the dominant domain for AI agent usage, with models showing significant investment and application in this area.
- Agents can operate within IDEs (like VS Code), command-line interfaces (CLI), and platforms like GitHub.
- Tools and extensions (e.g., AL language tools, MCP servers) enable agents to understand, generate, and interact with AL code, including debugging and setting breakpoints.
This section emphasizes the transformative impact of AI agents on the software development lifecycle, enabling developers to be more productive and focus on higher-level tasks.
An agent in VS Code can be instructed to 'set a breakpoint where we validate if a certificate is assigned to an item,' and it will find the correct code location and set the breakpoint automatically.
- The BCBench framework is used to evaluate the performance of coding agents in Business Central, measuring bug fixes and test generation.
- This framework allows for systematic testing of models, tools, and instructions, using real-world commit data.
- The process involves a 'hill climbing' strategy: test changes, keep improvements, revert regressions.
- Key findings include that more context isn't always better; precise instructions can significantly improve agent accuracy.
Understanding evaluation frameworks like BCBench is crucial for developing and improving AI agents, ensuring they are effective and reliable rather than counterproductive.
An experiment showed that providing overly long instructions to an agent designed for writing tests actually decreased its performance by 5%, while shortening and refining the instructions improved accuracy.
- The Expense Agent is a dedicated application (web app and mobile app) for managing employee expenses.
- It can process receipts in various languages, extract key information (amount, date, merchant), and categorize expenses.
- The agent can also handle complex tasks like itemizing hotel bills and calculating per diem allowances based on travel details.
- Users do not need a Business Central license to use the expense agent, making it accessible to a wider range of employees.
This provides a concrete example of how AI agents can streamline a common, often tedious, business process, improving user experience and operational efficiency.
An employee can forward a hotel bill to the expense agent, which then automatically itemizes the charges and categorizes it, saving the employee significant manual data entry.
Key takeaways
- AI agents represent a significant leap from earlier AI features, moving from narrow assistance to autonomous end-to-end process management.
- The rapid advancement of AI models, measured by the 'time horizon' concept, is the foundational technology enabling more capable agents.
- Business Central is integrating AI agents natively and through partner solutions, allowing for customized automation within the platform.
- Building custom agents involves defining their capabilities (seeing, permissions) and providing clear instructions, with an iterative design and testing process.
- External platforms like CoPilot Studio can integrate with Business Central via MCP connectors, expanding the possibilities for agent creation and deployment.
- AI agents are revolutionizing software engineering by assisting with coding, debugging, and testing, making developers more productive.
- Rigorous evaluation frameworks like BCBench are essential for measuring and improving the performance of AI agents, ensuring they deliver tangible benefits.
- Dedicated AI agents, like the Expense Agent, can simplify complex business processes, offering accessible and efficient solutions for end-users.
Key terms
AI AgentsCoPilotTime HorizonAgentic AIMCP Server (Model Context Protocol)MCP ConnectorAgent Design ExperienceSDK (Software Development Kit)Engineering AgentsBCBenchExpense Agent
Test your understanding
- How does the concept of 'time horizon' explain the difference in capabilities between early AI features like CoPilot and modern AI agents?
- What are the key components that define an AI agent's functionality, and how are these configured within Business Central?
- Explain how an external tool like CoPilot Studio can interact with Business Central data using the MCP connector.
- Describe the role of engineering agents in software development and provide an example of how they can be used in an IDE like VS Code.
- Why is an evaluation framework like BCBench important for the development and deployment of AI agents?