AI Agents as Workforce: What Microsoft and Deloitte Data Means for IT Governance

The prediction: AI gets a seat on the workforce agenda​

Angela Vidler, managing director and co-founder of Hyvid, wrote the piece for IT Brief Australia. Her central claim is that 2027 is when AI agents move from the technology agenda onto the workforce agenda. Boards currently govern people through headcount, pay, skills, succession and structure. She expects them to ask the same questions of agents:

• What is the agent responsible for?
• Who owns it?
• How is it performing?
• What decisions can it make?
• What data can it access?
• Who is accountable when it gets something wrong?

Her example is a company with 500 employees whose real capacity is a mix of people, AI-augmented people and autonomous systems doing defined work. Headcount alone would not show what that company can deliver.

She also predicts that:

• Businesses will move from asking where to use AI to asking how work should be redesigned now that AI can do part of it.
• Global and offshore teams will be judged on output per person rather than lowest cost. She expects one AI-enabled person to own a much larger volume of work. She gives no figures or examples for this.
• Workforce planning will shift toward capacity, meaning how much work is needed and what mix of people and AI delivers it.

Her piece is a forecast. "2027 will be the year" is her opinion, not a finding. Hyvid sells offshore and global talent, so she has a commercial interest in a future where global teams are valued for capability rather than cost. That is not disqualifying, but readers should weigh it.

What Microsoft's Work Trend Index actually says​

Vidler leans on Microsoft's 2026 Work Trend Index. Microsoft published the report on 5 May 2026, and it argues that organisations, not workers or technology, are the bottleneck.

The 15x figure. Microsoft says the number of active agents in the Microsoft 365 ecosystem grew 15x year over year, and 18x in large enterprises. The definition matters:

• The count covers unique active agents in the Microsoft 365 Copilot Agents platform and SharePoint in a rolling 28-day period.
• An agent counts as active if it had at least one day of user-initiated use in that window, or completed at least one autonomous run.
• The data spans March 2025 through March 2026.
• The agent creation platforms included are Agent Builder and Microsoft 365 Agents Toolkit.

This is a growth ratio inside Microsoft's own ecosystem. It is not a count of all agents, and it does not show that agents improved productivity. Several outside commentators raised the same caveat. GeekWire noted that Microsoft did not disclose the baseline, which makes the actual scale of adoption hard to judge. Fifteen times a small number is still a small number.

The 67% versus 32% split. Vidler says organisational factors account for "more than twice" the reported AI impact of individual factors. Microsoft's figures are 67% organisational and 32% individual, covering culture, manager support and talent practices. Microsoft states that these are statistical associations, not causal effects. Every variable is self-reported by the same respondent at the same moment. The sample was 20,000 AI-using knowledge workers across 10 markets, including Australia, with fieldwork run by Edelman Data x Intelligence. The survey ran from February 18 to April 7, 2026. Microsoft gives a slightly longer window, to April 20, for its AI-impact and readiness analyses.

Frontier Professionals. Microsoft's most advanced AI users are 3,233 of the 20,000 respondents. Microsoft defines them by self-reported behaviour: advanced agent use for multi-step work, routine workflow redesign, and structured, repeatable practices. Compared with other respondents, they were more likely to say their organisations document agent workflows, human handoffs and quality standards:

These are descriptive comparisons. They do not prove that documentation alone produces better outcomes. The report also finds that only about one in five respondents (19%) sits in the "Frontier" zone, where individual skill and organisational readiness are both high.

Deloitte: high expectations, low readiness​

The second pillar is Deloitte's survey, published 12 August 2026. It covers 501 senior manager-to-C-suite respondents in the US across five industries, fielded April to June 2026. All respondents' organisations were at least piloting agentic AI. The sample is therefore not representative of business in general, and it is US-only.

The headline numbers Vidler cites:

• 74% of leaders expect nearly half of their business processes to be redesigned or rebuilt around AI agents within four years. (The wording comes from Deloitte's release.)
• 61% expect most agents they use to be generally autonomous, with humans overseeing.
• 75% say human-agent collaboration creates more value than agent automation alone.
• Only 5% say their business processes are highly prepared for agents.

Deloitte's release has more detail, and Vidler's piece leaves much of it out. The survey measures several different things, and they should not be mixed up:

• 21% of leaders say they are prepared to redesign processes for autonomous operation. Business processes are the least-prepared of seven areas.
• 15% have scaled, orchestrated, cross-functional multi-agent adoption.
• 31% expect most processes to be redesigned around agents within two years, which is much lower than the four-year figure.
• 43% expect significant workforce disruption within 12 to 18 months, and half say their organisations are not investing enough in workforce transformation.
• The barriers cited were lack of unified data (72%), inability to trust and govern agents (70%), and costly, complex integration (67%).

These are respondents' expectations and opinions. They are not measured outcomes. The 5% figure is "highly prepared" only. It should not be read as the share of companies that are prepared in any sense.

Why this matters to Windows and Microsoft 365 admins​

Vidler's board questions map closely onto ordinary IT administration. Microsoft's report says as much. For IT leaders, it says agents should be treated as managed entities with identities, permissions, policy enforcement and lifecycle management, with IT acting as the control plane. For security teams, it names three risks: data exfiltration, unintended system actions and unauthorised access. It recommends building monitoring, policy enforcement and auditability into the platform.

That is a practical checklist. Based on the questions in both reports, an admin team could start here:

1. Inventory. Know which agents exist, who built them and which platforms they run on, such as Copilot agents, SharePoint agents and custom builds.
2. Ownership. Assign a named human owner to each agent. Microsoft's report asks who reviews agent performance and who may change the workflows agents run.
3. Access. Record what data each agent can reach, and review that access on the same schedule as user permissions.
4. Decision rights. Define what an agent may do on its own and where a human must approve.
5. Evaluation. Decide how output quality is checked. Microsoft warns that approving one bad output is manageable, but bad outputs that slip through at scale compound the risk.
6. Lifecycle. Retire agents that are no longer used, and plan for role changes among their owners.

This list is general guidance drawn from the questions in the two reports. It is not an official Microsoft procedure. Neither report prescribes specific admin-centre settings, so check Microsoft's own documentation for your tenant.

Balanced view​

The "agents as workforce" framing has fair critics. It is easy to dress up a software deployment as a workforce, and a board that treats a scheduled automation like an employee may add paperwork without adding control. The data behind the story also comes from interested parties. Microsoft sells Copilot, and some analysts note that the Work Trend Index is vendor research. Hyvid sells global talent. Deloitte sells transformation consulting.

Still, the independent pieces point the same way. Both surveys describe organisations that adopt agents faster than they redesign the processes around them. The risk they identify is governance, not model quality.

The bottom line​

• Vidler's 2027 predictions about boards, offshore talent and capacity-based planning are forecasts. They are not established trends.
• Microsoft's 15x and 18x growth in active agents is real telemetry, but it has no disclosed baseline and does not measure productivity.
• The 67% versus 32% split is an association from self-reported survey data, not proof of cause.
• Deloitte's 5% figure means "highly prepared" only. The broader picture is one of strong expectations, limited scaling and weak process readiness.
• For IT admins, the most actionable idea is to manage agents like any other identity. Give them an owner, scoped access, audit trails and a retirement date.