The distinction depended on whether the system could act

Gartner forecast on 26 August that 40 per cent of enterprise applications could include task-specific AI agents by the end of 2026, up from less than five per cent in 2025. The prediction did not measure realised value. More useful was its distinction between an assistant responding to a user and an agent completing a bounded, multistep task. Gartner described the practice of confusing the two as agentwashing.

In a customer journey, an assistant might propose campaign copy or summarise a case. A task-specific agent could select an audience within rules, begin an approved conversation, qualify a lead or create a service activity. The more an agent acted, the more consent, data quality, financial and channel limits, and a clear point for human judgement mattered.

The Dynamics roadmap showed where boundaries met

Microsoft's July wave 2 plan described agents and Copilot across Customer Insights, Sales, Customer Service, Contact Center and Field Service. Customer profiles and orchestrated journeys would provide signals to sellers. Case and knowledge agents would support service, while the field plan included agents, scheduling, mobile work and a connection to Project Operations. The plan covered October 2025–March 2026 and represented intended delivery rather than August availability.

The cross-functional opportunity was also the main process risk. A marketing signal could become a sales lead, service case or field visit. Without a shared customer identity and status, several agents could contact the same person, rely on stale consent or create duplicates. Process owners therefore needed to define which systems could write each field and which agent could close, escalate or reopen work.

Gartner's Hype Cycle on 5 August placed both AI agents and AI-ready data at the peak of inflated expectations. It recommended tight connection with a business need, infrastructure assessment and collaboration between business and AI teams. That perspective moderated the vendor's broad roadmap: an agent should be selected because a particular handover could improve, rather than because every customer application had gained agent features.

The process map became the agent's contract

An organisation could begin with one task, such as following up an agreed service reminder. The agent contract needed a valid trigger, permitted data sources, channel and quiet times, allowed actions, financial limits and the moment a person took over. The team could then measure incorrect recipients, duplicate activities, complete handovers, correction time and customer response. These findings provided a better basis for the next agent than a general count of generated messages.

Sources

Gartner: Task-specific AI agents in enterprise applications, 26 August 2025.

Microsoft Dynamics 365 Blog: 2025 release wave 2 plans, 16 July 2025.

Gartner: Hype Cycle identifies top AI innovations in 2025, 5 August 2025.

Microsoft Dynamics 365 Blog: AI-powered proactive engagement in public preview, 3 June 2025.

For discussion

Which bounded customer task has a clear trigger, permitted actions and a measurable endpoint?