Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Microsoft is building Work IQ for agent-first enterprises.
- Agents can discover data structures dynamically at runtime.
- The biggest concerns are cost, governance, and exposure.
Work IQ is a new offering from Microsoft that showcases two quintessential Microsoft skills: the ability to solve complex technical and infrastructure problems with an elegantly sophisticated solution, and the ability to make something almost impossible to explain. But I’m going to try.
Work IQ is the result of Microsoft completely redesigning how enterprise software works. Yeah, it’s that big.
If you think about how the enterprise software ecosystem has worked for the past few decades, it’s consisted of applications and data (together, let’s call them “solutions”) that worked on their own, or passed data between one another.
Also: Enterprise AI agents are multiplying fast, and Microsoft wants full control of them
Those solutions were often linked either through data transfer protocols or APIs. But no matter what, some human had to code the connection between any two solutions. Integrating something new into the mix, therefore, took a great deal of coordination, development, integration, and, let’s be honest, meetings. Oh, so many meetings.
But Microsoft is betting that 2026 marks the transition point between a human-driven enterprise world and an AI-agent-driven one. Microsoft describes it as, “Work IQ is built for an agent-first world, where AI agents — not human developers — decide in real time which tools to use across systems.”
Also: These top 30 AI agents deliver a mix of functions and autonomy
This opens up so many questions. Stay tuned to the end of this article, because I had the chance to ask Bryan Goode, Microsoft corporate vice president of Business Applications and Agents, about some specific concerns that I immediately thought of when reviewing this announcement.
But first, let’s understand as best we can just what it is Microsoft is doing to change, well, everything about enterprise IT operations.
Paradigm shift
Let’s look at this from a line-of-business perspective. Let’s say you’re an executive at a clothing manufacturer. Suddenly, retailers are starting to see a flood of returns from a previously successful product. Examination of the returns doesn’t raise any red flags: the clothes are in pristine condition and have no odd smells.
Finding the answer to this mystery using traditional enterprise software could be nigh on impossible. You might assign a person or team to investigate the problem, but the API-linked enterprise software might not help surface any problems.
Also: I set 10 honesty traps for Claude Opus 4.8 – and a legal test broke it
But now, let’s assume you have an agent-first software ecosystem. You ask an agent to solve the mystery. The agent and a series of sub-agents cross-reference SKU return rates, logistics routing maps, and customer service complaint keywords (like “itchy,” “rash,” or “sneezing”).
The agent eventually returns one common factor: every returned item spent at least 48 hours in Bay 4 of Warehouse A7. It turns out that Bay 5 was used to store materials for industrial adhesives and microscopic chemical residue was finding its way into the fibers of the clothing in Bay 4.
Traditional IT systems manage pre-established linkages of factors, relations that are consciously cross-linked by database developers. Agents have to be able to query everything in the enterprise, sift through all that information, and aggregate it all into an answer.
Also: True agentic AI is years away – here’s why and how we get there
This kind of system can be incredibly powerful and helpful. But Microsoft has apparently figured out that traditional IT infrastructure can’t get there from here.
Adaptable enterprise intelligence
There’s a lot to Work IQ, but two key elements spotlight the new agentic approach to data. The first is a capability called getSchema, which “allows agents to dynamically discover how data is structured at runtime. Instead of relying on predefined models or integrations, agents can understand what data exists, how it’s organized, and how to interact with it as needed.”
In other words, an agent can ask a data structure, “tell me about yourself,” and the data structure will do just that.
Also: 5 ways to fortify your network against the new speed of AI attacks
This means that agents don’t have to hold the entire data context of the enterprise at once. AI works using a context window (think of it as something like short-term memory). If that context window attempts to grow too big, the AI forgets some of what’s there, resulting in inaccuracies and the industry’s most disturbing euphemism: hallucination.
Operationally, the agent might start with a table of resources and ask each resource to describe the information it holds. If the resource answers in a way that the agent finds intriguing, then the next step is to do something with that data.
As with everything else about traditional enterprise software, “do something with that data” is often a very big job, requiring understanding of thousands of possible operations in the enterprise. Work IQ is designed to offer “a compact, efficient interface that minimizes context and can adapt as requirements change.”
Also: Euro-Office, Europe’s open-source alternative to Microsoft Office, launches June 9
Microsoft says it has “collapsed” thousands of operations into just 10 generic tools. These tools provide access to Microsoft 365 data and provide the mechanisms to work on that data. Those tools have simple functions like fetch, create, and update. The functions, like the tools and getSchema itself, are standardized across the organization. This makes it possible for the agents to construct a dynamic set of operations in real time.
Here’s how Microsoft explains it: “By exposing structure on demand, Work IQ turns every data source into a self-describing interface. This allows agents to adapt automatically to new data and evolving scenarios without changes to the API surface.”
But what about Copilot?
Oh, Copilot is still very much in the mix. Think of Copilot as the living space in a house, and Work IQ as the plumbing. You use the sink. The sink itself uses the plumbing to deliver and remove water.
Microsoft is adding Ask APIs that “expose the full M365 Copilot Chat experience to external applications as a single, opaque service. For every query, the system internally handles reasoning, tool selection, and action execution, delivering the same depth, context, and intelligence as Copilot.”
Then Work IQ mixes in custom instructions and saved memories to tune responses to the way users wish to be interacted with. Over time, as memories build up, the user can ask follow-up questions without having to repeat entire conversations.
The tough questions
There are obvious governance, budgeting, and security issues involved in all of this. Rather than regurgitating points from Microsoft’s press release, I had the opportunity to pose some very pointed questions to corporate VP Bryan Goode. Let’s dig in.
ZDNET: What evidence do we have that Work IQ-powered agents will produce durable savings or revenue gains, not just another layer of licensing, integration, monitoring, and support?
Microsoft: Work IQ APIs are optimized for the unique needs of agents in the workplace, which access data and tools in very different ways than a human.
Also: Why enterprise AI agents could become the ultimate insider threat
An agent-native API allows us to deliver higher quality results for agentic use cases; an optimized retrieval system that enables fewer round trips to the service resulting in lower latency and better token efficiency; higher scale data access and throughput; and processing that stays within the tenant boundary. All this adds up to agents that are higher quality, faster, more secure and less expensive.
ZDNET: Before we talk architecture, APIs, agents, or governance, what specific business result are we expecting that we cannot get through ordinary automation, better search, better reporting, or existing Microsoft 365 Copilot features?
Microsoft: AI agents access data and use tools very differently than humans. Relying on traditional APIs and connectors means that agents end up with lower-quality results, slower performance and higher cost — not to mention the potential for security and compliance issues if data is moved outside the tenant boundary.
Work IQ APIs are built for the unique needs of agents and address each of those challenges, while expanding the usability of Work IQ beyond Microsoft 365 to any agent or service that customers are building.
ZDNET: Are we creating a new centralized intelligence layer that attackers, insiders, compromised accounts, or misconfigured agents can exploit? Does the promised runtime “choke point” reduce risk, or does it become a high-value target?
Microsoft: Any concentrated capability is a target, but the alternative is worse: every agent standing up its own data store, with its own data movement, its own auth, its own audit gaps. Centralizing in Work IQ shrinks the surface area. Data, context, and insights stay inside the tenant trust boundary.
Also: When old data brings AI rollouts to a screeching halt – and how to manage it
Every call is authenticated through Microsoft Entra, including the new Entra Agent ID for non-human identities, and trimmed to what the signed-in user is already allowed to see. Every action is auditable and discoverable in Purview and Agent 365 alongside the rest of the Microsoft 365 estate.
ZDNET: The document says pricing is consumption-based and tied to tool calls, orchestration, and reasoning. What prevents a badly designed agent, a runaway workflow, or normal enterprise scale from producing unpredictable cost?
Microsoft: Consumption pricing means that customers pay only for what they use; it also means that we need to deliver FinOps capabilities that help customers manage that spend effectively.
With the general availability of the Work IQ APIs, we’ll also introduce new consumption management capabilities in the Microsoft 365 Admin Center that allow admins to set tenant, group, and user-level cost controls, create notification triggers, and monitor usage.
ZDNET: The document says Work IQ is built for an agent-first world, with agents choosing tools and acting across systems. Does adopting this mean we need to redesign workflows, permissions, approvals, and operational controls around agents?
Microsoft: Work IQ is designed to extend the enterprise controls organizations already use in Microsoft 365 rather than introduce an entirely separate operating model for agents.
Existing permissions, identity, compliance, retention, DLP, auditing, and approval structures continue to apply because agents operate within the same tenant trust boundary and act in the context of authenticated users or managed agent identities.
What does change is that organizations are now capable of taking more autonomous, cross-system actions at much higher scale and speed. That means customers will naturally evolve some operational practices over time.
ZDNET: If Work IQ and Copilot memory retain user preferences, past chats, and saved memories, what becomes discoverable, auditable, removable, or governed under our existing policies?
Microsoft: Memory sits inside the customer’s tenant, governed the same way as the rest of their data. Users can see what’s been remembered about them, edit it, and delete it. Admins set retention and deletion policy.
It’s subject to Purview controls: eDiscovery, audit, DLP, sensitivity labels. It never leaves the tenant trust boundary, so the data subject rights customers already exercise under GDPR, HIPAA, and sector-specific regimes carry over. We deliberately avoided creating a new governance silo for AI.
Is this the future of IT? Should it be?
Microsoft is clearly betting that enterprise IT will be agent-first in the future. I’m not so sure. While there’s no doubt AI will have a huge impact, I’m wondering whether organizations are prepared for the huge costs associated with agentic AI usage on an enterprise scale.
Also: What you’ll pay for AI agents will be wildly variable and unpredictable
Already we’re seeing pushback against the cost of AI agents. It’s clear why agentic AI seems like a goldmine to the AI providers (and that includes Microsoft). If every single IT operation requires a substantial token consumption tax on top of already pricey IT investments, the productivity gains might not seem as attractive.
If you factor in the non-deterministic nature of AI operations, and then add in all of the security and governance operations hinted at above, agent-first starts to seem a bit more like a nightmare than a promise.
I have no doubt that agentic AI will make strong inroads into enterprise operations. The example I showed earlier of diagnosing a sales problem was just a minor example of the benefits. And Work IQ has a strong chance of helping make agentic AI palatable and workable for enterprise operations. In many ways, it’s groundbreaking.
Also: How Microsoft obliterated safety guardrails on popular AI models – with just one prompt
I just have my doubts that organizations are going to go all-in on agent-first operations. The costs are high, the switching process is enormous, and the downsides could be catastrophic. Instead, I expect we’ll see a hybrid approach, with some AI and some traditional operations.
After all, that’s how IT has always managed these big technological swings. Pilot programs. Small expansions. Larger expansions, and then a mix of legacy and new technology. Until the per-token costing issue is resolved, along with all the other gray areas AI seems to thrive in, hybrid will be the way of the future. Let’s not call it agent-first. Let’s call it agent-also.
Does the idea of AI agents choosing tools across enterprise systems sound like productivity magic or operational chaos? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
