Proposed Images:Building AI Agents on Azure: From Chatbots to Autonomous Workflows
It’s early Wednesday afternoon. Your IT director walks into the project room looking equal parts excited and rattled. "The procurement team just spent forty minutes on the phone chasing an invoice that was sitting in three different systems the whole time. Can we build something that just... handles that?"
Silence. Not the confused kind. The kind where everyone's thinking the same thing at once.
Because that "something" isn't a chatbot. It isn't a Copilot prompt either. It's an AI agent, a system that can reason through a problem, decide which tools to call, pull data from your ERP, cross-reference it against your finance platform, and take action without someone hovering over its shoulder at every step. And as of 2026, building one on Azure is no longer a research project. It's a trainable, certifiable skill.
The shift has been fast. PwC reported that eight in ten enterprises now use some form of agent-based AI. Azure AI Foundry is one such service and it now supports tens of thousands of organisations building agents through its Agent Service. But there's a gap between having the platform and knowing how to use it properly, which is where targeted training comes in.
The Difference That Actually Matters
Worth pausing on this, because the terminology gets murky.
A chatbot answers questions.
A Copilot assists with tasks when prompted.
An AI agent does something fundamentally different: it breaks a goal into steps, decides what tools to use, executes those steps across multiple systems, and adjusts course when something doesn't go to plan.
Think of it this way. A chatbot tells you the invoice is overdue. A Copilot drafts a follow-up email. An agent finds the invoice across your systems, checks the payment status in your finance platform, sends the reminder, updates the CRM record, and flags the account manager if the amount exceeds a threshold. Autonomously.
That's not a subtle distinction.
Three building blocks sit underneath every agent:
A language model provides the reasoning engine.
Tools (APIs, databases, enterprise connectors) give the agent the ability to act on the world rather than just describe it.
And orchestration logic ties those pieces together, handling multi-step planning, error recovery, and the decision-making that determines what happens next.
On Azure, Microsoft Foundry Agent Service manages the hosting, scaling, identity and enterprise security so your team can focus on the agent logic itself.
The interesting thing is how quickly the tooling has matured. Twelve months ago, building a multi-agent system required stitching together open-source frameworks with custom glue code. Now you've got the Microsoft Agent Framework (which merges AutoGen and Semantic Kernel into a unified SDK), native support for the Model Context Protocol (MCP), and built-in tracing, evaluation and red-teaming tools. The platform is genuinely production-ready. The question is whether your team has the skills to use it.
Where Organisations Get Stuck
We've seen this pattern play out a few times now.
The pilot works. Production doesn't. A developer builds a proof-of-concept agent in a notebook. It's impressive. But moving it into a governed, secure, scalable environment requires architecture decisions that weren't part of the prototype. Identity management. Network isolation. Tool authentication. Content safety guardrails. The gap between "demo" and "deployed" is wider than most teams expect.
Nobody owns the architecture. Development teams build agents. Security teams evaluate them. Platform teams deploy them. But who's responsible for the end-to-end design? For deciding whether you need a single agent or a multi-agent system? For working out how agents interact with your existing Dynamics 365 or Power Platform investments? That architectural role is new, and most organisations haven't filled it yet.
The skills gap sits in two different places. Your developers need hands-on capability with the Azure AI Agent Service and Semantic Kernel. Your architects and senior consultants need to understand agentic design patterns, governance frameworks, and enterprise integration at a strategic level. Those are different training needs, and trying to address both with a single course doesn't work.
Azure AI Courses That Build Agent Capability
Three courses map directly to the skills gap. They cover different roles and different levels of depth, so you can target training at the people who'll have the most impact.
1. AI-3026: Develop AI Agents on Azure
This is the starting point for anyone who'll actually build agents. One day, hands-on, and focused squarely on the practical skills your developers need.
Who it's for | Developers, data scientists and AI engineers building intelligent agent-based solutions on Azure |
Problem it solves | Your team wants to build AI agents but needs structured, practical training on Azure AI Agent Service and Semantic Kernel rather than figuring it out from documentation and blog posts |
Format | One day, hands-on. Leads to a Microsoft Applied Skills credential. |
Key skills | Azure AI Agent Service fundamentals, Semantic Kernel Agent Framework, tool integration, agent orchestration patterns, deploying agents for enterprise workloads |
One thing that makes AI-3026 particularly useful right now: it's a modular component of the AI-102: Azure AI Engineer Associate certification. So your team earns a standalone Applied Skills credential on day one, and that learning counts toward the full certification when they're ready. It's a flexible path that respects how busy technical teams actually learn. View the AI-3026 course.
2. AB-100: Architecting Agentic AI Business Solutions
Who it's for | Solution architects, enterprise architects, senior functional and technical consultants, and AI transformation leads responsible for designing agent-based systems across Microsoft platforms. This is explicitly not a developer course. It's for the people who need to make architecture and governance decisions before the developers write a single line of code. |
Problem it solves | Your organisation is planning agentic AI initiatives, but nobody has the architectural judgment to design solutions that are secure, governed, cost-effective and actually aligned with your existing Microsoft investments (Dynamics 365, Microsoft 365, Power Platform, Azure AI). |
Format | Instructor-led. Prepares for the Microsoft Certified: Agentic AI Business Solutions Architect credential. |
Key skills | Agentic design patterns, multi-agent orchestration, governance frameworks, cost-benefit analysis, lifecycle management, enterprise integration across Microsoft platforms |
The course focuses on design reasoning and trade-offs rather than step-by-step configuration. Think of it as the strategic counterpart to AI-3026. Where AI-3026 teaches your developers how to build agents, AB-100 teaches your architects how to design agent systems that will actually survive contact with enterprise reality: multi-agent orchestration patterns, lifecycle management, governance frameworks, and cost-benefit analysis.
Worth mentioning: this is a new certification for 2026. AB-100 prepares for the Microsoft Certified: Agentic AI Business Solutions Architect credential, which signals pretty clearly where Microsoft sees the enterprise skills market heading. View the AB-100 course.
3. AI-3016: Develop Generative AI Apps in Azure
If AI-3026 teaches your team to build agents specifically, AI-3016 covers the broader generative AI application layer that agents sit on top of. The two courses complement each other well, and for teams moving into custom Copilot and autonomous agent development, taking both makes practical sense.
Who it's for | Data scientists and AI engineers with existing knowledge of generative AI models and Python who want to build custom Copilots and generative AI applications on Azure AI Foundry |
Problem it solves | Your team understands AI concepts but needs practical experience building generative AI applications using Azure OpenAI Service, prompt engineering, and Azure AI Foundry's development tools |
Format | One day, hands-on. Leads to a Microsoft Applied Skills credential. |
Key skills | Azure AI Foundry platform, Azure OpenAI Service integration, prompt engineering for production apps, building custom Copilots, content filtering and safety controls |
Like AI-3026, this course stacks toward the full AI-102 certification. One day, one credential, real production skills. View the AI-3016 course.
What This Looks Like in Practice
OK so here's where it gets tangible. Agents aren't theoretical. Organisations across Australia and New Zealand are already deploying them in production, and the use cases are more grounded than you might expect.
IT service management. Atomicwork built an agent called Atom using Azure AI Agent Service that handles IT and HR requests autonomously. Early adopters reported measurable productivity gains and reduced operational costs. Not conceptual. Deployed and measured.
Sales automation. Fujitsu used Semantic Kernel and Azure AI Agent Service to build a multi-agent system for sales proposals. The result? A 67% improvement in proposal creation productivity, which freed sales teams to focus on actual customer engagement instead of document wrangling.
Quality assurance. JM Family deployed a multi-agent system with specialised agents for requirements gathering, story writing, coding, documentation and QA. Their development cycles went from weeks to days, cutting QA time by up to 60%.
The common thread across all of these? They're not single-task chatbots. They're systems of agents that coordinate, check each other's work, and manage complex multi-step workflows. That's the capability gap your team needs to close.
So Where Do You Actually Start?
Depends on what's holding you back.
If your developers are keen but building from blog posts and Stack Overflow threads, start with AI-3026. One day, structured, hands-on. They'll walk out with a working understanding of Azure AI Agent Service and Semantic Kernel, plus a credential that counts toward AI-102. That's immediate capability for anyone tasked with building agents.
If the bigger problem is that nobody's designed the system, AB-100 fills the architectural gap. Your architects and senior consultants get the design judgment they need to plan agentic solutions that integrate properly with your existing Microsoft stack and meet your governance requirements. Without this step, you end up with clever agents that don't talk to each other and can't pass a security review.
If your team needs to go broader across generative AI before specialising in agents, AI-3016 provides the foundational generative AI development skills that underpin agent work. Pair it with AI-3026 and you've got a two-day investment that covers the full spectrum from custom Copilots to autonomous agents.
Most organisations will benefit from a combination. Send your architects through AB-100 to sort out the design and governance layer. Send your developers through AI-3026 (and AI-3016 if they need the generative AI foundations). That gives you strategic clarity and hands-on capability at the same time, which is the only combination that actually ships working agents.
Learn to Build AI Agents on Azure with Lumify Work
Lumify Work delivers the broadest range of Microsoft AI and cloud training in Australia, New Zealand and the Philippines. As part of Lumify Group, we've skilled more people in Microsoft technologies than any other organisation in ANZ, and we hold the Microsoft MCT Superstars Award for FY24, recognising the highest quality Microsoft Certified Trainers in the region.
Training is available online, in-person across ten campuses, or as tailored programmes designed around your organisation's specific AI stack and team needs. Whether you need to get your architects across agentic design patterns with AB-100,, upskill your developers with AI-3026, or build generative AI capability with AI-3016, we've got you covered.
Explore Lumify Work's full AI and Machine Learning training catalogue and get your team building production-grade AI agents before the rest of the organisation moves without them.
