Enterprise agents graduate from answering to executing

From answering to resolving

Enterprise service management extends IT service management practices across the entire business, according to Serval, bringing teams such as IT and HR onto the same service model. The shift Serval describes is from routing requests to resolving them: AI-resolved requests complete work without human intervention, and Serval Journeys execute workflows deterministically once published. The determinism claim matters because an agent that improvises its way through an approval or provisioning step is a governance problem, not a productivity gain. m:…0245

The cross-departmental case is a request that would otherwise bounce between teams' queues, which Serval says automation can prevent. Serval allows IT and HR to run separate help desks with distinct workflows and permissions on the same platform, and its Catalyst tool allows admins to build journeys that coordinate work across teams. Serval supports just-in-time access provisioning for security governance, and the AI that builds workflows in Serval is separate from the agent that runs them, a separation that keeps design-time choices distinct from run-time actions. m:…0245

Serval also says it can run alongside existing help desk platforms like ServiceNow, rather than requiring a rip-and-replace migration. m:…0245

One connection standard instead of hundreds

The second Serval piece addresses the plumbing problem that grows as agents multiply. The Model Context Protocol reduces the repetition of building custom connections for AI assistants, and Anthropic introduced MCP as an open standard to prevent the need for building different connections for every AI application and tool pairing. An MCP server makes a system's capabilities available to an AI assistant through the protocol, with core components including tools, resources, and prompts. m:…4974

IT teams often support 5,000 employees across more than 40 systems, according to Serval, and point-to-point integrations become unmanageable at scale, potentially requiring 200 connections for 40 applications and 5 AI assistants. MCP allows teams to reuse connections as they add assistants and bring more systems into reach, enables faster time-to-value by allowing teams to connect to existing servers without writing new adapters, and allows different teams to use the same server even if they rely on different AI models. m:…4974

Serval says it has integrated a native MCP server into its ITSM platform, and that its integration respects caller permissions and credential scopes, with a consistent connection model for enforcing access rules across assistants. But the company is candid that MCP does not remove operational work: IT teams must still maintain servers, test tools, and address security risks like prompt injection when using MCP. A standard connection is not a solved security model, and any executive buying on the promise of effortless integration should read that caveat as part of the pitch. m:…4974

Chatbots that do the ticket, not deflect it

The third Serval piece describes enterprise chatbots evolving from simple ticket deflectors into autonomous agents that resolve requests end to end by executing code-based workflows. Enterprise chatbots resolve internal IT requests directly in Slack or Microsoft Teams, and complete tasks like granting access by running workflows connected to identity and ITSM systems. They handle routine tickets like software provisioning and password resets, and trigger actions in other apps through ITSM platforms, identity providers like Okta, and HRIS tools. m:…4666

The mechanism Serval describes is compilation, not conversation. Serval compiles plain language workflow descriptions into deterministic TypeScript, and admins can configure sensitive workflows to require human review before completion. Serval logs every agent action for auditability, according to the company. Those three features together, deterministic code, human review gates, and a full action log, are what separate an agent that executes from a chatbot that merely answers politely and creates a ticket for someone else. m:…4666

The piece also notes that platform selection should be matched to organizational structure and ITSM maturity, and that regulated industries require chatbots with deployment flexibility and customization for compliance. That is a useful restraint on the executing-agent story: the same autonomous resolution that helps a software company may be unacceptable in a regulated environment without deployment control and customization, and maturity, not enthusiasm, should set the pace. m:…4666

The value gap

Presight offers the counterweight to all of this tooling progress. It says 60% of companies globally are not generating material value from AI despite substantial investment, and contends that the industry obsession with technical metrics like model size and agent counts overshadows the mission of creating measurable impact. AI systems are moving from assistance and recommendations toward systems that act on behalf of users, it notes, which makes the value question more urgent, not less: an agent that acts badly at scale does more damage than a chatbot that answers badly. m:…2989

His prescription is architectural and organizational rather than technical. Future-ready AI architecture preserves the ability to change models or partners without rebuilding systems, governance and accountability must evolve alongside AI capability, and accountability for AI actions must remain with humans. Trust is a prerequisite for scaling innovation, he argues, and scale is the primary stress test for AI applications. AI should be judged by its ability to deliver existing organizational goals, and the most advanced technology today may not be the most appropriate for every use case tomorrow. m:…2989

Read against the Serval pieces, the tension is clear: the tooling for agents that execute deterministically, connect through a standard protocol, and log every action is arriving, largely from vendors describing their own platforms, while the value case measured in organizational goals still lags for most companies, according to Presight. Executives should ask vendors which existing goal an agent delivered, under whose accountability, and what happens when the model or partner changes. m:…0245 m:…4974 m:…2989