For the past few years, much of the AI conversation around Microsoft Dynamics 365 Business Central has focused on Copilot.
The first wave of capabilities largely helped users work faster, from generating product descriptions and analyzing data to assisting with everyday tasks. For partners, AI could therefore be viewed primarily as another productivity layer sitting on top of familiar Business Central processes.
That distinction is beginning to disappear.
Microsoft is moving Business Central beyond AI assistance towards agents capable of interacting with data, navigating processes, and completing multi-step tasks. At the same time, AI is changing how Business Central itself can be developed. GitHub Copilot agent tools can now build, publish, debug, and navigate AL projects, while Microsoft has introduced an AI development toolkit that allows partners to design and code their own Business Central agents.
This is more than an incremental change to the user experience. It has implications for solution architecture, permissions, integrations, extensions, development practices, and ultimately the technical capabilities Business Central partners need within their teams.
The opportunity is considerable. AI can reduce repetitive development work and automate increasingly complex business processes. But as software becomes easier to create, maintaining architectural discipline becomes more important, not less.
That opportunity is already translating into measurable operational gains. A 2026 Forrester Consulting study of Dynamics 365 Business Central projected up to 50%-time savings across accounts payable, accounts receivable and billing activities, alongside up to a 30% reduction in monthly financial close times by year three. As AI moves from assisting those processes to participating in them directly, the quality of the architecture supporting that automation becomes increasingly important
Business Central is moving from assistance to action
Copilot established the first phase of AI inside Business Central by helping users perform existing tasks more efficiently.
Agents change the model because they can potentially carry out work rather than simply assist someone carrying it out.
Microsoft’s Sales Order Agent illustrates that distinction. Rather than waiting for a user to manually process every incoming request, the agent can monitor a shared mailbox, identify customer enquiries, retrieve relevant information, create sales quotes or orders, and communicate with customers as the process progresses.
The Payables Agent applies a similar principle to finance. It can help process vendor invoices and work through the associated payable workflow.
The significance is not the individual scenarios. It is the architectural pattern behind them.
Business Central is evolving from a system in which AI primarily helps users interact with ERP processes towards one in which AI can participate in those processes.
Microsoft’s 2026 release wave 1 takes that further. Partners can now prototype agents directly within Business Central sandbox environments using an in-client design experience before implementing production-ready agents in AL. Business Central can also integrate with Microsoft Copilot Studio, while its Model Context Protocol (MCP) Server enables external AI clients to connect to Business Central data and business logic.
That opens a much wider set of possibilities for partners. Instead of waiting for Microsoft to deliver an agent for a particular process, they can begin designing AI-driven workflows around the specific requirements of customers and industries.
But it also changes the architectural questions they need to answer.
Agentic ERP requires a different approach to permissions and data
Giving an AI system access to ERP data is fundamentally different from giving a user access to a chatbot.
Business Central contains financial records, customer information, pricing, supplier data, inventory, transactions, and commercially sensitive business logic. Once agents can interact with those resources, identity and permissions become part of the AI architecture.
Microsoft’s Business Central MCP Server demonstrates why this matters. It allows AI clients to interact with Business Central environments and, depending on their permissions and configuration, perform activities such as viewing or editing records.
That means partners need to think carefully about what an agent should be able to see, which actions it should be permitted to perform, and where human approval remains necessary.
An agent capable of answering a question about inventory availability presents one level of risk. An agent capable of creating records, triggering workflows, or modifying financially significant information presents another.
The principle of least privilege therefore becomes critical. Agents should receive only the access required to complete their intended task, and sensitive actions need appropriate validation, auditability, and approval controls.
Data architecture matters too.
Agentic systems need context. If an agent is expected to reason across customers, orders, invoices, inventory, emails, and external applications, partners need to understand how those sources connect and which should be considered authoritative.
Integration design consequently becomes part of agent design.
Business Central’s MCP Server and Copilot Studio integration make it easier to expose Business Central data and processes to AI applications, while custom APIs exposed through AL extensions can make additional business logic available. The opportunity is to create much richer automation. The challenge is ensuring those integrations remain secure, understandable, and maintainable as the surrounding architecture expands.
AI is also changing how Business Central gets built
The other side of this transition is happening inside development teams themselves.
AI-assisted coding has moved quickly from autocomplete towards tools capable of reasoning about codebases and carrying out development tasks. Business Central is following that trajectory.
Microsoft introduced GitHub Copilot agent tools into the AL extension in 2025, allowing Copilot agent mode to automate workflows such as building AL projects, publishing extensions, navigating symbols, and starting debugging sessions. In May 2026, Microsoft went further by making the AL Language Server Protocol available outside Visual Studio Code, giving autonomous AI agents semantic understanding of AL projects, including references, dependencies, completions, and type hierarchies.
For experienced Business Central developers, these capabilities can remove considerable friction.
Boilerplate code can be produced faster. Developers can navigate unfamiliar projects more efficiently. Tests, documentation, debugging, and refactoring can increasingly be assisted by AI. Prototypes that previously required significant manual development can be explored much earlier in a project.
The wider software industry is already seeing this productivity effect. Google’s 2025 DORA research, based on nearly 5,000 technology professionals globally, found that 90% were using AI at work and more than 80% believed it had increased their productivity. The same research found that 59% reported a positive impact on code quality.
For Business Central partners facing growing delivery demands, those productivity gains are difficult to ignore.
But faster development creates its own architectural risk.
The technical debt risk behind 'vibe coding'
AI has dramatically lowered the effort required to generate working code.
That can be extremely valuable during prototyping, particularly when experienced developers use AI to accelerate tasks they already understand. Problems emerge when the speed of generation begins to replace architectural understanding.
The term ‘vibe coding’ has become shorthand for development in which someone describes the desired outcome to an AI system, accepts or iterates on the generated code, and focuses primarily on whether the application appears to work.
In a prototype, that may be enough.
In an ERP environment expected to remain secure, supportable, and upgradeable for years, it is not.
AI-generated AL can still introduce unnecessary dependencies, duplicate existing functionality, select inappropriate extension points, create inefficient queries, overlook permission requirements, or solve an immediate problem in a way that becomes difficult to maintain later.
The issue becomes more significant when AI increases the sheer volume of code teams can produce.
DORA’s 2025 research captures this tension particularly well. Despite the productivity benefits developers reported, 30% said they had little or no trust in AI-generated code. DORA also found that higher AI adoption was associated with increased software delivery throughput but continued to have a negative relationship with delivery stability.
Subsequent DORA research describes AI as a ‘mirror and a multiplier’: strong engineering environments can use it to move faster, while fragmented environments risk generating technical debt faster.
That is particularly relevant to Business Central.
Partners already need to manage the long-term consequences of customer-specific extensions and integrations across regular platform updates. If AI dramatically increases the speed at which custom functionality can be produced, teams need stronger mechanisms for deciding what should be built in the first place.
Maintainability becomes more important when development gets faster
The answer is not to restrict AI-assisted development. It is to surround it with stronger engineering practices.
Microsoft’s own direction reinforces this point.
The Business Central AI development toolkit allows teams to prototype agents without AL code, but Microsoft distinguishes that prototyping experience from productizing agents for customers. Production scenarios can be implemented in AL extensions, where teams can define setup, integrations, triggers, scheduled processes, and other functionality using established development and application lifecycle management practices.
The distinction is important.
AI can make experimentation dramatically faster without changing the requirements of production software.
Partners still need source control. They still need automated testing, code analysis, peer review, security controls, clear documentation, and disciplined application lifecycle management. Extensions still need to use appropriate interfaces and extension points. Integrations need to be resilient, and solutions need to survive future Business Central updates without creating unnecessary upgrade work.
AI arguably makes those disciplines more valuable because code generation is no longer the bottleneck it once was.
Google’s DORA research supports this. It found that AI can accelerate development but that organizations need robust control systems, including strong automated testing, mature version control practices, and fast feedback loops, to prevent increased development velocity from translating into instability. Among developers who write tests, 62% are already using AI to assist with testing, showing that AI can be part of the quality solution as well as part of the development challenge.
For Business Central partners, the objective should therefore be controlled acceleration: using AI to increase development capacity without sacrificing the qualities that make an ERP solution sustainable.
Business Central technical capability is evolving with the platform
This shift also changes what partners need from their people.
Traditional Business Central expertise remains essential. Consultants and developers still need to understand finance, supply chain, business processes, AL, extensions, integrations, security, and the underlying application architecture.
But AI introduces another layer.
Developers increasingly need to understand how to work effectively with AI-assisted development tools while still being able to evaluate the code those tools produce. Technical teams need to understand agent architecture, permissions, data access, APIs, MCP, Copilot Studio, and the governance implications of allowing AI systems to act against ERP data.
Functional consultants also have a role. Designing useful agents requires a detailed understanding of the business process being automated, the exceptions that occur within it, the information required at each stage, and where human judgement needs to remain.
This means AI capability cannot sit with one specialist at the edge of a Business Central practice. It increasingly needs to become part of how consultants, developers, and solution architects understand the platform.
Developing AI-ready Business Central teams
For partners, the challenge is developing those capabilities while continuing to deliver customer projects.
#BCTalent is designed to help partners build and develop Business Central capability through three flexible paths: ReSKILL, UpSKILL, and CrossSKILL. The initiative is already trusted by more than 200 Business Central partners, helping organizations bring new professionals into the ecosystem or develop the capabilities of people already within their teams.
ReSKILL helps transform professionals with relevant industry knowledge and transferable skills into Business Central specialists, combining product and professional skills training with optional recruitment support.
UpSKILL is designed for graduate and early career professionals, helping partners accelerate the development of Business Central and professional capabilities without placing the entire training burden on senior team members.
CrossSKILL enables professionals with existing Business Applications experience, including Dynamics GP and other ERP backgrounds, to transition their knowledge into Business Central.
As Business Central becomes more AI-enabled, these paths provide partners with a way to develop capability alongside the platform. Learners can access the latest AI skills as part of their development, while practical training and community sessions with Microsoft MVPs, product experts, and industry specialists help connect new platform capabilities with real Business Central delivery.
The objective is not to create AI specialists who lack ERP foundations. It is to develop Business Central professionals who understand the platform deeply enough to use AI responsibly and effectively within it.
That distinction will become increasingly important.
AI changes how Business Central is built, not what good architecture requires
Business Central’s progression from Copilot towards agents represents a significant evolution in what ERP systems can do.
Agents can automate processes that previously required continuous human interaction. AI-assisted development can shorten the distance between an idea and a working solution. MCP, Copilot Studio, and Business Central’s own AI development tooling give partners new ways to connect AI to real business data and processes.
But none of those capabilities removes the need for good architecture.
If anything, the opposite is true.
When software becomes easier to generate and ERP processes become easier to automate, partners need stronger judgement around permissions, integrations, extensions, testing, governance, and maintainability. The competitive advantage will not come from producing the greatest volume of AI-generated customization. It will come from knowing where AI creates value, how to architect it safely, and when not to customize at all.
The next generation of Business Central professionals will therefore need to combine platform knowledge with AI fluency and strong engineering judgement.
For partners, developing that capability now is an opportunity to benefit from faster development and more intelligent automation without creating the technical debt that could undermine those gains later.