Follow ZDNET: Add us as a preferred source on Google.
Highlights from ZDNET
- Microsoft is developing Work IQ for organizations built around AI agents.
- AI agents can explore and understand data setups on their own, as they go.
- The main worries involve expenses, oversight, and security.
Work IQ is a brand-new tool from Microsoft. It shows off two typical Microsoft strengths: solving tough technical and setup problems with a smart, polished fix, and making something very tricky to explain. Still, I’ll give it a shot.
Work IQ comes from Microsoft’s total overhaul of how business software is built. Yes, it’s a major change.
For years, the business software world has centered on apps and data—let’s bundle them together as “solutions.” These either worked alone or shared data with each other.
Also: AI agents in businesses are growing quickly, and Microsoft aims to manage them all
These solutions usually connected through data-sharing methods or APIs. Either way, someone always had to write the code to link two systems together. Bringing something new into the fold, then, needed lots of planning, coding, integrating, and—let’s face it—meetings. So many meetings.
But Microsoft believes 2026 is when things shift: from a world run by humans to one run by AI. According to Microsoft, “Work IQ is designed for a world led by AI, where AI helpers—not human coders—pick tools across systems in the moment.”
Also: Microsoft’s MDASH rolls out with over 100 focused AI agents for threat detection
This idea brings up so many questions. Stick around until the end of this piece because I spoke with Bryan Goode from Microsoft, who leads Business Applications and Agents, about some issues that jumped to mind when I read the news.
Before that, let’s break down what Microsoft is doing—which really means changing nearly all of how business tech works.
Big Change Ahead
Let’s think from a business side. Imagine you work for a clothing brand. Suddenly, stores start sending back tons of a once-popular product. Looking at the returns, everything seems fine: clothes look new and don’t smell odd.
Finding out why with old business software was nearly impossible. You could give someone or a team the job, but standard linked apps might not catch the issue.
Also: I tested 10 honesty challenges on Claude Opus 4.8—and one legal question cracked it
Now, picture a workplace built around AI agents. You tell an AI to figure it out. The AI and smaller helper agents check return rates per item, shipping routes, and common words in complaints (like “itchy,” “rash,” or “sneezing”).
After checking, the AI finds a pattern: each returned piece sat for at least 48 hours in Bay 4 of Warehouse A7. It turns out Bay 5 was used for industrial glue supplies, and tiny traces of chemicals had soaked into clothes stored in Bay 4.
Old-school IT links data that developers planned and set up ahead of time. Now, AI must search all business info, look through everything, and piece together an answer.
Also: Microsoft introduces its first reasoning AI among seven new tools revealed at Build
This kind of setup can be super helpful. But Microsoft says current business systems just can’t handle this leap.
Smarter, Flexible Business Tools
Work IQ has a lot of parts, but two main ones stand out for the new AI-driven data style. The first is getSchema, which “lets AI discover how data is set up right when it’s needed. Rather than using fixed plans or built-in links, AI can learn what data is there, how it’s grouped, and how to use it as they go.”
In simple terms, AI can ask any data file, “Who are you?” and the system happily explains itself.
Also: 5 steps to strengthen your network against faster AI-powered attacks
This way, AI doesn’t need to grasp everything in the company at once. AI uses something like a short-term memory, called a “context window.” If that window grows too big, the AI misses pieces, leading to mistakes and that unnerving term: hallucination.
In action, an AI might start with a list of resources, asking each what kind of details it holds. If something sounds useful, the AI works with that info next.
Like the rest of old business software, “working with data” used to be huge, requiring knowledge of endless company tasks. Work IQ offers “a slim, smooth way to interact that cuts clutter and adjusts as needs shift.”
Also: Euro-Office, Europe’s free alternative to Microsoft Office, arrives June 9
Microsoft says it’s narrowed thousands of tasks down to just 10 simple tools. These work with Microsoft 365 data and handle everything inside. Each tool does basic jobs like fetch, create, or update. These tools, like getSchema, are all built the same across the business, so AI can build custom tasks instantly.
Here’s how Microsoft puts it: “By showing structure only when needed, Work IQ turns every data spot into a system that describes itself. This means AI adapts on its own as data changes or situations shift, no extra API work needed.”
So, Where Does Copilot Fit In?
Copilot’s still very much part of the picture. Think of Copilot as your main living area at home and Work IQ as the pipes behind the walls. You use the faucet. The faucet taps into those pipes to get water in and out.
Microsoft is rolling out Ask APIs that “give outside apps full access to M365 Copilot Chat, all in one wrapped-up service. Each request sent to it is handled inside the system—thinking, picking tools, and doing actions—just as smartly as Copilot would.”
Then Work IQ adds saved preferences and memories so replies can match the way each user likes to talk. Over time,as memories accumulate, users can pose follow-up questions without needing to rehash entire past discussions.
The tough questions
There are clear governance, budgeting, and security challenges inherent in all of this. Instead of merely repeating points from Microsoft’s press release, I had the chance to ask corporate VP Bryan Goode some very direct questions. Let’s dive in.
ZDNET: What proof exists that Work IQ-powered agents will generate lasting savings or revenue increases, rather than just adding another tier of licensing, integration, monitoring, and support?
Microsoft: Work IQ APIs are specifically tailored for the distinct requirements of workplace agents, which interact with data and tools in a fundamentally different manner than humans.
Also: Why enterprise AI agents could become the ultimate insider threat
An agent-native API enables us to provide superior outcomes for agentic scenarios; a streamlined retrieval system that minimizes round trips to the service, leading to reduced latency and improved token efficiency; greater scale for data access and throughput; and processing that remains within the tenant boundary. Together, this results in agents that are superior, quicker, more secure, and more cost-effective.
ZDNET: Before discussing architecture, APIs, agents, or governance, what precise business outcome are we anticipating that cannot be achieved through standard automation, improved search, enhanced reporting, or current Microsoft 365 Copilot capabilities?
Microsoft: AI agents retrieve data and utilize tools quite differently than people. Depending on conventional APIs and connectors leads to subpar results, sluggish performance, and elevated expenses — not to mention the possible security and compliance risks if data is transferred outside the tenant boundary.
Work IQ APIs are constructed for the specific demands of agents and resolve each of those problems, while also extending the utility of Work IQ beyond Microsoft 365 to any agent or service that customers are developing.
ZDNET: Are we establishing a new centralized intelligence tier that hackers, malicious insiders, hijacked accounts, or poorly configured agents could take advantage of? Does the anticipated runtime “choke point” lower risk, or does it simply turn into a high-value bullseye?
Microsoft: Any concentrated capability is a target, but the alternative is far worse: each agent setting up its own data repository, managing its own data transfers, handling its own authentication, and creating its own audit blind spots. Centralizing within Work IQ minimizes the surface area. Data, context, and insights remain securely inside the tenant trust boundary.
Also: When old data brings AI rollouts to a screeching halt – and how to manage it
Every request is verified via Microsoft Entra, incorporating the new Entra Agent ID for non-human identities, and restricted to what the logged-in user is already permitted to view. Every action is auditable and traceable in Purview and Agent 365 alongside the rest of the Microsoft 365 environment.
ZDNET: The documentation states that pricing is based on consumption and linked to tool calls, orchestration, and reasoning. What stops a poorly built agent, an uncontrolled workflow, or standard enterprise scaling from resulting in unforeseen expenses?
Microsoft: Consumption-based pricing indicates that customers only pay for their actual usage; it also implies that we must offer FinOps tools to assist customers in handling that spending efficiently.
Alongside the general release of the Work IQ APIs, we will also roll out new consumption management features in the Microsoft 365 Admin Center. These will empower admins to enforce tenant, group, and user-level budget limits, establish alert triggers, and keep an eye on usage.
ZDNET: The documentation claims Work IQ is engineered for an agent-first environment, where agents select tools and execute actions across various systems. Does adopting this require us to restructure workflows, permissions, approvals, and operational safeguards around agents?
Microsoft: Work IQ is crafted to augment the enterprise controls that organizations already employ within Microsoft 365, rather than forcing a completely distinct operational framework for agents.
Current permissions, identity, compliance, retention, DLP, auditing, and approval setups remain valid because agents function within the identical tenant trust boundary and operate under the context of verified users or managed agent identities.
Also: Apple, Google, and Microsoft join Anthropic’s Project Glasswing to defend world’s most critical software
What does shift, however, is that organizations can now execute more independent, cross-system actions at a significantly greater scale and pace. Consequently, customers will organically adapt certain operational practices as time goes on.
ZDNET: If Work IQ and Copilot memory store user preferences, previous conversations, and saved memories, what becomes searchable, auditable, erasable, or controlled under our current policies?
Microsoft: Memory resides within the customer’s tenant, regulated identically to all their other data. Users have the ability to view what has been retained about them, modify it, and erase it. Administrators determine retention and deletion rules.
It falls under Purview restrictions: eDiscovery, auditing, DLP, and sensitivity labels. It never departs the tenant trust boundary, meaning the data subject rights that customers currently utilize under GDPR, HIPAA, and other sector-specific regulations remain applicable. We purposefully avoided establishing a fresh governance silo specifically for AI.
Is this the future of IT? Should it be?
Microsoft is undoubtedly wagering that enterprise IT will become agent-first in the years ahead. I remain unconvinced. While AI is guaranteed to make a massive impact, I question whether businesses are truly ready for the hefty price tags tied to agentic AI deployment on an enterprise scale.
Also: What you’ll pay for AI agents will be wildly variable and unpredictable
We are already observing resistance regarding the expense of AI agents. It’s obvious why agentic AI appears to be a lucrative opportunity for AI vendors (Microsoft included). If every single IT procedure demands a substantial token consumption tax piled onto already expensive IT investments, the productivity benefits might lose their appeal.
When you consider the unpredictable nature of AI processes, and then layer on the security and governance efforts referenced earlier, an agent-first paradigm begins to look more like a potential disaster than a guarantee.
I am confident that agentic AI will significantly penetrate enterprise workflows. The earlier illustration of identifying a sales issue was merely a small glimpse of the advantages. Furthermore, Work IQ stands a good chance of making agentic AI acceptable and practical for corporate operations. In many respects, it’s revolutionary.
Also: How Microsoft obliterated safety guardrails on popular AI models – with just one prompt
I simply harbor doubts that companies will fully commit to an agent-first operational model. The financial burdens are steep, the transition is massive, and the potential drawbacks could be disastrous. Instead, I anticipate a blended methodology, combining AI with traditional operations.
After all, this is how IT has historically navigated these major technological shifts. Pilot projects. Gradual expansions. Broader rollouts, followed by a mixture of legacy and modern technology. Until the per-token pricing dilemma is settled, alongside the myriad other ambiguities where AI tends to flourish, a hybrid model will represent the path forward. Rather than calling it agent-first, let’s refer to it as agent-also.
Does the concept of AI agents picking tools across corporate systems sound like a productivity miracle or operational mayhem? Share your thoughts in the comments section 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.
