AI Intelligence
Live on the platformMove from questions to coordinated tasks—with clear human control.
Task-focused assistants with bounded tools and human oversight
AI Agents & Copilots are task-oriented assistants that help users understand information, prepare work and coordinate permitted actions across DISHA 4.0 HCOS™. A copilot typically supports a person while they remain in control. An agent may plan and execute a bounded sequence of steps using approved tools. Both require scoped context, transparent behavior, reliable handoffs and safeguards against unauthorized actions.
Explore assistants designed to reduce repetitive work while keeping decisions and consequential actions in human hands.
Governed tool-calling agents run in the AI Command Center today.
In depth
Copilots for human-led work
A copilot can help explain a career pathway, summarize a profile, draft a development plan, prepare interview practice or navigate a module. The user reviews and edits the output. The experience should identify AI-generated content and provide a route to inspect supporting information where available.
Agents for bounded workflows
An agent can break a task into steps, call authorized tools, check results and report completion. Examples may include assembling a learning shortlist, preparing a profile-completion checklist or organizing a draft workforce scenario. Each agent needs a defined objective, tool allowlist, permission scope, budget or step limit, and stopping condition.
Context and memory boundaries
Only the minimum user-authorized profile context should be passed to an assistant. Workspace and tenant boundaries must remain enforced. Persistent memory should be explicit, inspectable and editable where provided. The system must not silently reuse another person’s data or infer sensitive traits.
Human approval for consequential actions
Actions such as submitting an application, contacting a person, changing shared records, enrolling in a paid service or publishing information require explicit confirmation. The interface should show the action, destination, data involved and expected consequence before execution.
Reliability and evaluation
Agent workflows should be tested for tool correctness, instruction conflicts, prompt injection, partial failures, duplicate actions and recovery. The agent must distinguish a completed action from a draft, a failed tool call or an unverified suggestion.
How it works
Capability domains
Frequently asked questions
Can an agent take actions for me?
Only within configured permissions. Consequential or external actions should pause for explicit confirmation.
Can I review what an assistant did?
The product should provide a proportionate task summary and, where supported, a trace of tools and outcomes.
AI Intelligence
Experience it inside the live platform.
Open the module that carries this capability, or talk it through with the team.