Microsoft Just Split Agentic AI Into Two New Certifications: What AB-100 and AB-620 Actually Test

Microsoft quietly added two brand new exam codes to its certification catalog on September 4, 2026: AB-100 and AB-620. Neither one is a rename of something older, and neither replaces an existing exam directly. They mark the point where Microsoft stopped treating agentic AI as a feature bolted onto existing Azure and Copilot certifications and started building it a certification track of its own.

That distinction matters more than it sounds. Microsoft’s older AI credentials, AI-900 and AI-102, were built around a world where you called pre-built AI services from your code. AB-100 and AB-620 are built around a different world: one where AI agents plan tasks, call tools, talk to other agents, and run inside production workflows with minimal human babysitting. Job postings for that world grew 280% year over year according to Stanford’s 2026 AI Index, and Microsoft’s certification team clearly noticed before most training providers caught up.

This piece breaks down what each exam actually tests, what it costs, how the two connect to Microsoft’s broader AI certification reset, and whether it is actually worth sitting either one this early in their life cycle.

Why Microsoft Just Split Agentic AI Into Two Separate Certifications

Building an AI agent and deciding what an AI agent should be allowed to do are not the same skill, and Microsoft’s new track treats them as two different jobs rather than two halves of one job.

• The business and architecture side (AB-100): planning which processes should get an agent, designing the governance and security guardrails around it, and deciding how it fits the organization’s existing systems.
• The technical building side (AB-620): actually configuring, integrating, testing, and shipping the agent inside Copilot Studio and connecting it to enterprise data and other systems.

Plenty of certifications claim to cover “strategy and implementation” in a single exam and end up shallow on both halves. Microsoft’s approach here is narrower on purpose. A solutions architect who never touches Copilot Studio can still sit AB-100. A developer who never writes a business case can still sit AB-620. That split is itself the news, since it tells you Microsoft expects two distinct job titles to form around agentic AI rather than one generalist role.

Inside AB-100: The Agentic AI Business Solutions Architect Exam

AB-100 leads to the Microsoft Certified: Agentic AI Business Solutions Architect Expert credential, and “Expert” is not a marketing label here. It sits at the top of Microsoft’s role-based ladder, above the associate-level prerequisites it requires.

Exam Format, Cost, and Passing Score

• 100 minutes, delivered through Pearson VUE
• 40 to 60 multiple-choice and scenario-based questions
• Passing score of 700 out of 1000
• Exam fee of $165 USD (prices vary by country and local tax)
• Domain split: planning AI-powered solutions (25 to 30%), designing them (25 to 30%), and deploying them (40 to 45%)

The content leans heavily on Microsoft Copilot Studio, Microsoft Foundry, the Model Context Protocol, and Agent-to-Agent protocol work, plus the governance and monitoring questions you would expect from an exam that sits at the architecture level rather than the code level.

Why AB-730 or AB-731 Comes First

Microsoft does not let candidates walk straight into AB-100. You need one of two associate-level prerequisites already in hand:

1. AB-730, AI Business Professional, or
2. AB-731, AI Transformation Leader

Both prerequisite exams cost $99 USD, roughly $66 less than AB-100 itself, and both are considerably shorter and more conceptual. In practice this means the real cost of reaching AB-100 for someone starting from zero is closer to $264 across two exams, not the $165 sticker price on AB-100 alone. That two-exam stacking pattern is becoming common across Microsoft’s newer role-based tracks, and it is worth budgeting for before you commit to a study plan built around a single exam fee.

Who Actually Needs This Credential

AB-100 fits solutions architects, enterprise architects, and senior consultants who are being asked to decide where agentic AI belongs in an organization, not necessarily build it themselves. If your actual day job is writing Copilot Studio flows and wiring up integrations, AB-620 is the closer match, covered next.

Inside AB-620: The AI Agent Builder Associate Exam

AB-620, officially Designing and Building Integrated AI Agent Solutions in Copilot Studio, leads to the Microsoft Certified: AI Agent Builder Associate credential. Despite being filed under Microsoft’s “app maker” exam family on paper, the actual content skews toward developers and consultants integrating Copilot Studio into enterprise systems rather than citizen developers doing light no-code work.

Exam Format, Cost, and Passing Score

• 120 minutes, delivered through Pearson VUE
• Passing score of 700 out of 1000
• Exam fee of $165 USD
• No formal prerequisite exam required, unlike AB-100
• 93 questions on DirectCertify’s current practice set, covering single choice, multiple choice, hotspot, and drag-and-drop formats

No formal prerequisite does not mean no real barrier. The exam guide expects working familiarity with Power Fx, Microsoft Dataverse, the Power Platform, Microsoft 365 Copilot, Microsoft Foundry, adaptive cards, and enough generative AI background to talk comfortably about retrieval-augmented generation and the Model Context Protocol. Someone who has never opened Copilot Studio will find the 120 minutes go by fast.

The One Domain That Decides Whether You Pass

Integrating and extending agents makes up 40 to 45% of AB-620, the single largest domain on the exam. That section leans on Azure AI Search, Microsoft Foundry, Microsoft Fabric, agent-to-agent protocol behavior, and Power Platform application lifecycle management pipelines. Candidates who have actually shipped a Copilot Studio agent into a production integration describe this section as a fair reflection of the work. Candidates relying purely on reading tend to get caught out here specifically, since it rewards hands-on building time more than memorized definitions.

Who Should Take This Instead of AB-100

AB-620 is the right target for IT application developers, consultants, and ISV partners who are the ones actually building and shipping agents, not deciding organizational strategy around them. If your resume already leans developer rather than architect, this is very likely the certification your manager is thinking of when they say “get certified in the new AI agent stuff.”

How AB-100 and AB-620 Fit Into Microsoft’s Bigger AI Certification Reset

AB-100 and AB-620 did not appear in isolation. Microsoft retired the long-standing AI-102 exam and its Azure AI Engineer Associate credential on June 30, 2026, replacing it with AI-103, Developing AI Apps and Agents on Azure, built around Microsoft Foundry rather than individual pre-built AI services. Anyone who already covered what AI-102 used to test will notice AI-103 shifts the entire center of gravity toward generative AI application development and multi-agent orchestration, the same direction AB-100 and AB-620 push from a different angle.

Put the current ladder side by side and the strategy becomes obvious:

Notice that AB-620 and AI-103 sit at the same associate level but cover different tools: AI-103 is code-first Azure AI Foundry development, while AB-620 is Copilot Studio and Power Platform-first. Microsoft is not asking developers to pick one over the other so much as expecting different developer profiles to land on different sides of that line depending on which platform their employer has already standardized on.

The Job Market Numbers Behind the Timing

Microsoft rarely builds a certification track this fast without labor market data behind it. A few numbers explain the urgency:

• Agentic AI’s share of US job postings grew from 0.06% to 0.23% of all listings between 2024 and 2025, a 280% year-over-year jump, according to Stanford’s 2026 AI Index.
• That translates to roughly 90,000 US job listings mentioning agentic AI skills in 2025 alone, a category that barely existed in postings two years earlier.
• ZipRecruiter’s July 2026 data puts the average AI Solutions Architect salary at $145,963 a year, with the middle 50% of earners between $126,000 and $166,000 and the top 10% clearing $183,000.

Those figures track a role, not a certification, so holding AB-100 or AB-620 does not guarantee any specific salary. What the numbers do show is that employers are already writing job descriptions around agentic AI work faster than most certification bodies have historically moved, which is likely why Microsoft shipped two exams in the same week rather than one combined exam months later.

Should You Actually Sit AB-100 or AB-620 Right Now?

A brand new exam code is not automatically worth your money the day it goes live. A few things are worth weighing before you book a seat:

• Practice assessment lag: Microsoft’s own practice assessments for a new exam typically become available within about eight weeks of general availability, so early sitters are studying mostly from the skills outline and community-written guides rather than an official practice test.
• Fast revision cycles: DirectCertify’s own AB-620 practice set is already on version 6.3 and AB-100 on version 10.2, both updated as of September 4, 2026, which signals content is still being tightened this soon after launch. Expect the skills outline to keep shifting for a few months.
• Prerequisite cost stacking: if you do not already hold AB-730 or AB-731, budget for that exam first before AB-100, not as an afterthought.
• Job description match: check whether the roles you are actually applying for name Copilot Studio and agent building (lean AB-620) or AI strategy and governance (lean AB-100), rather than assuming either credential covers both.

If your employer has already committed to Copilot Studio for internal automation, sitting AB-620 now puts you ahead of a very thin talent pool. If you are further from the actual keyboard and closer to deciding where agents belong in the business, AB-100 is the better use of study time, provided you already have AB-730 or AB-731 sorted out.

Microsoft is not the only vendor restructuring certifications around agentic work this year. Anthropic built an entire Claude certification program from scratch in 2026, and UiPath split its own automation credential into separate associate and professional agentic automation exams the same year. The pattern across vendors is consistent: agentic AI is getting its own certification lane rather than staying a chapter inside older exams.

One honest note before the FAQ below: DirectCertify is an independent certification prep provider and has no affiliation with, endorsement from, or sponsorship by Microsoft. AB-100 and AB-620 are Microsoft’s own trademarked credentials, and their content outlines, prerequisites, and pricing are still moving targets this early in the exams’ life cycle. Confirm current registration details, prerequisites, and exact pricing directly on Microsoft’s own certification pages before you book a seat.

Common Questions About Microsoft’s New Agentic AI Certifications