On June 16, Microsoft announced the global general availability (GA) of Copilot Cowork.
Around the same day, Anthropic announced strategic partnerships with two of the world's top IT service providers, TCS and DXC Technology, with the Claude model being deeply integrated into the service delivery systems of these two companies.
I initially thought this was a coincidence, but after carefully examining the timeline and context—it's not a coincidence. The common signal from both events is: Enterprise AI is evolving from an "assistant" to a "colleague," and from "tool procurement" to "outsourced service".
This change has a deeper impact on CIOs and digital leaders than the technology itself.
First, let's talk about Microsoft Copilot Cowork
In March, Copilot Cowork was still a Frontier preview, and on June 16, it went directly to global GA. This speed is quite fast for Microsoft.
What is the biggest difference between Copilot Cowork and the previous Copilot? It doesn't answer your questions; it does the work for you.
Let me translate:
Previous Copilot: You ask "Help me write a meeting minutes", it writes a draft, you copy it to Outlook.
Cowork: You say "Help me organize the minutes of this morning's client meeting and distribute them," and it automatically monitors the meeting, generates the minutes, sends them to attendees and absent colleagues, and adds them to the CRM.
This is no longer "AI helps you improve efficiency," but "AI has become a colleague at the workstation."
Key data released by Microsoft:
During the three-month preview period, more than half of the Fortune 500 companies have already adopted Copilot Cowork.
Major clients such as Accenture, Avanade, and Advance Local have already used Cowork in core processes such as financial reconciliation, compliance review, customer service, and sales lead processing.
What does this mean? It means that Copilot Cowork is no longer a "pilot tool" but a "formal employee" of large enterprises—complete with performance indicators, compliance reviews, data trails, and billing systems.
Along with this GA, Microsoft also did three things:
- Metering and billing: Charged based on "Agent task execution volume", no longer by annual seat packages
- IT Governance Tools: Administrators can audit the input/output, permission scope, and operation logs of each Agent
- Permission isolation: Data is not shared between Agents of different departments or different businesses
None of these three things is sexy, but all are indispensable—they address the compliance, billing, and auditing issues that inevitably arise when "AI Agents enter core processes."
I especially want to mention billing:
Microsoft has shifted from "charging per seat" to "charging per Agent task volume." This change has a significant impact on enterprises' IT budgets.
Previously, when you purchased 100 Copilot seats, the cost was fixed. Now, you buy 100 seats with a variable cost based on task volume, and when the month-end settlement arrives, you discover that the Agent automatically ran 5,000 tasks overnight, making the bill 40% higher than expected.
The CIO must plan the budget pool and trigger cap in advance for this matter; otherwise, once AI agents enter the company, "AI bill runaway" will become a new pain point in the second half of 2026.
Let's talk about Anthropic's two big deals
On June 11, Anthropic signed two world-leading IT service providers on the same day:
- TCS (Tata Consultancy Services): Top 5 global IT outsourcing company, with over 600,000 employees
- DXC Technology: An IT services giant spun off from HP, with an annual revenue of $13 billion
Both companies will embed Claude (especially Fable 5 and Mythos 5, just released on June 10) into their own service delivery systems.
What does this mean?
Global customers of TCS and DXC can now use Claude for their business by default. Banks purchasing TCS's IT outsourcing services automatically gain Claude's coding capabilities; insurance companies using DXC's claims processing system automatically gain Claude's document processing capabilities.
Previously, when enterprises adopted AI, they had to purchase models, find integrators, and handle data governance themselves. Now, outsourcing service providers directly offer packaged solutions, shifting the procurement model of enterprise AI from "project-based" to "service subscription-based".
The true meaning of this is:
AI agents are becoming the "utilities" of enterprise IT services.
When you buy a house, you don't negotiate "how to get tap water into the house" separately; you bundle it with the property management fees.
AI Agent is also following this trend — in the future, companies will no longer separately bid for "deploying an AI Agent," but rather "AI capabilities are a default configuration in the ERP/CRM/IT outsourcing I purchase."
Anthropic didn't choose TCS and DXC casually this time. These two are the main IT outsourcing forces in "highly regulated industries" — banking, insurance, healthcare, and government. Getting Claude into these industries is ten times harder than getting it into the internet industry.
Anthropic dares to sign, indicating they are confident in their model's compliance, safety, and interpretability. Fable 5 achieved the highest score (80.3%) on code capability tests like SWE-bench, but more critically, it passed security reviews in these highly regulated industries.
Two Things Viewed Together: The "Watershed" of Enterprise AI
Looking at these two events separately, each is news. Putting them together, I see a watershed moment:
June 2026 is the watershed moment when enterprise AI shifts from "tool" to "colleague".
There are three signs of switching:
| Dimension | Previously: AI tools | Now: AI Colleague |
|---|---|---|
| Procurement Method | The enterprise buys and deploys it themselves | Outsourcing service provider bundling, subscription model |
| Billing method | Per seat/year | By Agent task volume, by result |
| Assessment Method | User satisfaction, time savings | Business results, KPI completion |
| Governance approach | IT department usage guidelines | Company-wide data governance + compliance review |
| Enterprise Role | AI is a tool, employees use it | AI is an employee, the company manages |
This table was organized based on the characteristics of these two matters.
The last line in red — Enterprise Role, has changed the most. In the past, IT managed "tools" and business departments "used" them; now IT must "manage" AI Agent permissions, billing, compliance, and performance evaluation, while business departments need to interact with AI as "colleagues with defined roles and responsibilities."
This is not a matter of opinion, but a matter of organization.
Practical Suggestions for Enterprise CIOs/Digital Leaders
Seeing this, you shouldn't be asking "Should we adopt AI Agent?"—while you're thinking about it, your competitors are already doing it.
You should ask three questions:
Question 1: Has the billing method for our company's AI budget changed?
If you still pay by seat on an annual basis, in the second half of 2026 you will most likely see actual costs exceed the budget by over 40%.
It is recommended to do two things immediately: set a monthly trigger cap and plan an AI task volume billing pool.
Question 2: Are AI capabilities a default configuration in our ERP/CRM/IT outsourcing contracts?
Outsourcers like TCS, DXC, Accenture, and IBM have basically added AI capability clauses to their 2026 contracts. If you signed a contract last year that is still in effect, quickly negotiate a supplementary agreement with the supplier to explicitly include AI capabilities—including model selection rights, billing transparency, and data security boundaries.
Question 3: Have we written the "job description" for the AI Agent?
Joking—but not entirely joking.
Alongside the GA of Copilot Cowork, Microsoft specifically emphasized one point: Each Agent's permissions, input/output, and exception handling must have clear documentation, otherwise IT governance cannot manage it.
The same applies when integrating into an enterprise: you must clearly define for every deployed Agent "what it can do, what it cannot do, and who is responsible if something goes wrong." Otherwise, the Agent might make a mistake in the middle of the night, and you won't find out until you come to work the next day.
At the end, I want to share a judgment:
The wave in June 2026, is not a technological change, but a change in procurement model.
The capabilities of AI Agents were essentially in place by Q4 2025. But what truly enabled them to "run at scale within enterprises" were billing models, governance tools, and outsourced packaging—these unremarkable yet indispensable "engineering capabilities".
The winners of this wave of enterprise AI will not be the companies with the strongest models, but those that "package AI Agents into basic services like water, electricity, and gas."
Microsoft is fighting this way, Anthropic is fighting this way, and IBM, ServiceNow, SAP, and Salesforce are also fighting this way.
As a corporate CIO, you don’t need to bet on the right model—what you should bet on is "who can turn AI Agents into my infrastructure."
This is the most worthwhile thing to think carefully about in enterprise digitalization in the second half of 2026.
