The Digital Colleague: How AI Agents Join the Team

In a matter of weeks this autumn, four of the largest names in enterprise software presented AI agents that are meant to work less like tools and more like members of staff. Salesforce listed its Agentforce Coworker as generally available, Microsoft introduced Autopilot as a long-running agent with an identity of its own, Cisco invited agents into Webex meetings, and Google gave its new coworker agents their own email addresses. Behind the vocabulary is a practical shift: companies are moving from testing agents to employing them, and they are building the identity, oversight and measurement that any new colleague requires.

The adoption figures explain why the language has changed. In KPMG’s quarterly AI Pulse survey of 314 U.S. executives at companies with at least $1 billion in revenue, conducted between late July and late August, 62 percent said they are building, deploying or developing AI agents, up from 53 percent one quarter earlier. Twenty-five percent reported multi-agent systems, in which several agents divide a task between them; in each of the two preceding quarters, that share was 6 percent. Forty-four percent reported significant adoption of AI by their employees, compared with 23 percent in the previous quarter and 10 percent a year earlier, and nearly six in ten reported measurable business value.

McKinsey’s State of AI survey, published in August, points in the same direction. Among companies with more than $1 billion in revenue, 40 percent of respondents said their organizations are scaling AI agents, up from 27 percent a year earlier, and 80 percent of all 1,719 participants from 97 countries said AI had improved their individual productivity. Employees report the change as well: in Gallup’s survey of 22,573 U.S. workers in May, 47 percent said their organization had integrated AI tools, six points more than one quarter before, and 52 percent said they use AI in their role.

From assistant to teammate

The product announcements of September and October share a design idea: an agent is no longer something an employee opens, but something that works alongside them, sometimes for days at a time. On September 25, Microsoft introduced a new version of Copilot that brings chat and Cowork, for delegated end-to-end tasks, together in one place, along with Code for building solutions and Autopilot, a long-running agent formerly known as Scout. Autopilot lives in the customer’s own tenant with its own identity, memory, computer and workspace, and Microsoft describes it as a digital teammate. It enters private preview at the end of October; Cowork, Code and Autopilot are billed by usage, alongside the per-user license.

Salesforce’s Dreamforce announcements listed Agentforce Coworker as generally available. According to the company, 100,000 users activated it within its first 35 days. The Adecco Group, the staffing company, is rolling out Coworker, which runs on Anthropic’s Claude, to 27,000 employees in more than 40 countries after a pilot in the U.K. and France. Salesforce also announced long-horizon agents that pursue a goal over days, weeks or months; they are in pilot now, with general availability planned for November.

Cisco followed on October 7 at WebexOne. Teams can now invite AI agents into Webex spaces, meetings and calls to carry out multistep work across Webex and other applications, with guardrails from Cisco AI Defense and telemetry from Splunk; Cisco calls them digital teammates. A day later, Google introduced its Gemini agent at Gemini at Work 2026. Its coworker agents receive an identity of their own: an email address on an agents subdomain of the company’s domain, persistent storage and a Workspace account with email, calendar, Drive and an entry in the company directory. They run on Gemini models and on Anthropic’s Claude models.

An identity of its own

The detail that an agent receives its own account matters more than it might seem. In most companies, access to systems is organized around people: an employee gets an account, a role, a set of permissions and a manager who approves them, and every action can be traced back to a name. An agent that acts for many employees, or for a process rather than a person, fits into that model once it has an identity of its own.

Google records every action of a coworker agent in an audit trail under the agent’s name rather than a person’s, and lets customers set a hard spending cap per project. Microsoft’s Autopilot follows the same logic inside the customer’s tenant. The scale of the task shows in Microsoft’s own figures: on its July earnings call, the company reported nearly 40 million agents registered in Agent 365 across tens of thousands of companies, two months after its launch, and more than 30 million paid Microsoft 365 Copilot seats.

A category of software has formed around the same need. In October, Dataiku made Agent Management generally available, a product that finds every AI agent a company runs, whichever platform it was built on – AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce, Snowflake Cortex, or custom setups connected through OpenTelemetry. The principle resembles the onboarding of a new hire: an account, defined permissions, a responsible owner and a record of what was done.

Governance moves up the org chart

As agents take on more work, the question of what an agent may decide on its own has become a leadership topic. In KPMG’s U.S. survey, 49 percent of companies have defined high-risk use cases in which agents may not make decisions on their own, and 73 percent said they are confident in their AI governance, up from 57 percent one quarter earlier. KPMG’s global edition, based on 2,131 senior leaders in 20 countries, found that 53 percent place accountability for AI-informed decisions at the C-suite level or above, and that 34 percent see significant adoption of AI agents by their employees, up from 25 percent in the first quarter. The same leaders plan to invest an average of $210 million in AI over the next twelve months, compared with $186 million in the first quarter.

Deloitte’s survey of 501 U.S. executives, all at companies at least piloting agentic AI, shows how far the redesign is expected to reach: 42 percent have tested or deployed AI agents, 15 percent have scaled, orchestrated multi-agent adoption, and 31 percent expect most of their processes to be redesigned around agentic AI within two years. Because a digital colleague touches both roles and systems, the work involves HR as much as IT. In a Gartner survey published on October 7, 38 percent of chief human resources officers and chief information officers said they share an understanding of how AI affects the future of work.

Customer service sets the benchmark

No function measures the work of agents as closely as customer service, where every conversation has an outcome. Gartner found in August that AI spending by customer service leaders rose 38 percent, while overall service budgets grew by 2 percent. Within two years, the 199 service leaders surveyed expect generative AI chatbots to be their most valuable service channel.

Customer expectations are equally well documented. In Gartner’s survey of 3,566 customers, 50 percent said interactions are easier when companies use generative AI, and 58 percent of generative AI users have already used it to complete a task on their behalf – 74 percent in business-to-business settings. Eighty-seven percent said that companies using generative AI for service should offer a way to reach a human agent. The design that follows from these figures combines both: the agent resolves what it can and hands the rest, with its context, to a person.

Vendors now report results per conversation. On its own help site, Salesforce says Agentforce has handled 4.3 million inquiries and resolved 70 percent of them; the help agent it sells to customers is priced per resolution. Fin, the company formerly known as Intercom, whose acquisition for about $3.6 billion Salesforce completed on September 10, serves more than 30,000 business customers across chat, email, WhatsApp, SMS, voice and Slack. Among the customer results Salesforce published in September, Engine’s agent Eva fully resolves half of all chat inquiries, Autism Queensland’s agent Paige resolves 70 percent of administrative requests, and Fin autonomously resolves 79 percent of the Anthropic conversations it sees. All of these figures are reported by the company.

New measures for a new kind of work

As agents move from conversations to tasks, the units of measurement change as well. Salesforce now reports “Agentic Work Units”: 7.0 billion delivered to date, 3.2 billion of them in the second quarter of its fiscal year 2027, 97 percent more than in the quarter before. Its annual recurring revenue from Agentforce, which since that quarter also includes Slackbot and Headless 360, exceeded $1.5 billion, more than 240 percent above the previous year. Google reported that nearly 500 Google Cloud customers each processed more than one trillion tokens over the past year.

The tools for managing agents follow the same path. Salesforce made Agent Health Monitoring generally available with more than 16 built-in metrics and is testing a Task Resolution metric for voice and conversation agents. Costs are tracked just as closely: Google lets customers cap spending per project, Microsoft bills its agents by usage, and in McKinsey’s survey about one in five respondents said AI operating costs, including tokens, were a factor in how widely they use AI.

Company-reported results give a sense of scale. Google cited Tata Steel, which built more than 300 agents in nine months and halved the turnaround time for complaints; Commerzbank, which shortened a document review from 20 hours to one; and NTT DOCOMO, which returns 450,000 hours a year to the business. Microsoft’s Work Trend Index found that the number of active agents in Microsoft 365 grew fifteenfold in a year, and 66 percent of the 20,000 AI users surveyed said AI gives them more time for high-value work.

Human roles change alongside. In a Gartner survey of service leaders published in April, 85 percent said they are adding tasks and responsibilities to the roles of their human agents, and 75 percent are moving agents into new roles. At Adecco, candidates rated the first wave of recruitment agents in the U.K. 4.6 out of 5, and 57 percent of the agents’ conversations took place outside standard working hours.

Open standards as shared infrastructure

Agents that work across systems need common ways to reach tools and each other, and much of that infrastructure now has a neutral home. The Linux Foundation formed the Agentic AI Foundation in December 2025 with three founding projects: the Model Context Protocol, contributed by Anthropic; goose, from Block; and AGENTS.md, from OpenAI. Its platinum members are AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. In August, the foundation welcomed 57 new members, among them Alibaba, Visa and Wells Fargo, bringing the total to 247 organizations, and took in A2A, the open protocol for communication between agents, whose version 1.0 shipped in March and which more than 150 organizations support.

The protocols are already part of everyday products. Microsoft’s Dynamics 365 exposes more than 650,000 actions through MCP, Atlassian connects ChatGPT and Codex to its tools through an MCP plugin, and Cisco offers a Webex MCP integration. The July release of the MCP specification made the protocol stateless, and in September the Linux Foundation took in TRACE, a specification from AMD, Anthropic, Intel, Microsoft, NVIDIA, OPAQUE and TII that provides verifiable runtime and compliance evidence for AI agents.

What the coming weeks bring

The calendar keeps the subject in view. Gartner’s IT Symposium runs from October 19 to 22, followed on October 22 and 23 by AGNTCon + MCPCon North America in San Jose, where more than 3,500 attendees are expected. Microsoft’s Autopilot enters private preview at the end of October, and in November Salesforce plans general availability of its long-horizon agents, while Cisco ships RoomOS 27.

Gartner expects 40 percent of enterprise applications to include task-specific AI agents by the end of 2026, up from less than 5 percent in 2025. The companies now giving their agents accounts, permissions, owners and metrics are laying the groundwork for that – and treating the agent as what the new products call it: a colleague.

---

Sources and further reading

• Google Cloud, Thomas Kurian: Welcome to Gemini at Work 2026: Introducing the Gemini agent, October 8, 2026
• Cisco: Cisco Unveils New Agentic Collaboration Experiences, October 7, 2026
• Gartner: Gartner Survey Finds Only 38% of CHROs and CIOs Share an Understanding of How AI Impacts the Future of Work, October 7, 2026
• Atlassian: Grounding Frontier Intelligence in Enterprise Context, October 6, 2026
• Microsoft, Jared Spataro: Introducing the new Copilot with Home, Code and Autopilot, September 25, 2026
• Salesforce: 21 Things We Announced at Dreamforce 2026, September 28, 2026
• KPMG U.S.: AI’s Value Story Sharpens as Organizations Gain Confidence in Governance, Accountability and Workforce Adoption, September 24, 2026
• KPMG International: New KPMG AI Pulse Survey: As AI maturity converges, leading organizations show what AI at scale requires, September 24, 2026
• Dataiku: Dataiku Launches Agent Management to Track and Manage Performance of AI Agents Built and Running on Leading Platforms, September 24, 2026
• Linux Foundation: Linux Foundation Newsletter: September 2026, September 17, 2026
• Salesforce: Salesforce Unveils AIforce, Bringing the Full Power of Its Platform to Any Interface, September 15, 2026
• The Adecco Group: Adecco Group announces Salesforce’s Agentforce Coworker rollout across 40 countries, September 15, 2026
• Salesforce: Salesforce Expands Agentforce With a New Portfolio of AI Agents Built for High-Value Work, September 11, 2026
• Salesforce: Salesforce Completes Acquisition of Fin, September 10, 2026
• Gartner: Gartner Survey Finds AI Spending by Customer Service Leaders Has Surged by 38%, Despite Overall Service and Support Function Budgets Rising by Just 2%, August 26, 2026
• Salesforce: Salesforce Delivers Record Second Quarter Fiscal 2027 Results, August 26, 2026
• McKinsey & Company: The state of AI in 2026: On the road to ROI, August 25, 2026
• Agentic AI Foundation: A2A joins AAIF’s open agentic stack, August 17, 2026
• Linux Foundation: Agentic AI Foundation Welcomes 57 New Members, Gaining Major Financial Services Players and APAC Leaders, August 13, 2026
• Deloitte: AI Agents are Only the Beginning: Deloitte Survey Examines the AI Readiness Gap and Reveals How Enterprises Can Prepare for Agentic Success, August 12, 2026
• Gartner: Gartner Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent, August 4, 2026
• Microsoft Investor Relations: Microsoft Fiscal Year 2026 Fourth Quarter Earnings Conference Call, July 29, 2026
• Gallup, Andy Kemp: Organizational AI Adoption Jumps Six Points, July 20, 2026
• Salesforce: Salesforce Launches Agentforce Help Agent That Deploys in Minutes and Only Charges for Resolutions, June 25, 2026
• Microsoft WorkLab, Work Trend Index 2026: Agents, human agency, and the opportunity for every organization, May 5, 2026
• Gartner: Gartner Survey Finds 85% of Service and Support Leaders are Expanding Human Agent Responsibilities Despite Expectations of Mass AI Layoffs, April 28, 2026
• Linux Foundation: Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md, December 9, 2025
• Gartner: Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025, August 26, 2025
• The Adecco Group: The Adecco Group completes successful first agentic AI implementation at scale with Salesforce’s Agentforce, July 30, 2025