ActiveNudge - Autonomous Employee Onboarding Agent

Inspiration

Employee onboarding is often treated as a checklist, but behind that checklist are multiple people, systems, approvals, and dependencies.

One small delay can block everything.

A common example is a hiring manager who receives a form asking for a new employee's laptop, software, and access requirements but doesn't complete it on time. The onboarding workflow remains stuck, while IT teams manually follow up through emails or messages.

We asked:

What if an AI agent didn't just remind people about a blocked workflow, but actually resolved the bottleneck?

That idea became Active Nudge.

Our goal is to move enterprise onboarding from passive reminders to proactive resolution.

What It Does

Active Nudge is an autonomous employee onboarding agent that detects stalled onboarding workflows, proactively contacts the responsible manager, understands natural-language responses, and unblocks the workflow.

Imagine Rahul is joining as a Software Engineer.

His onboarding is stuck because his manager hasn't provided his requirements.

Instead of sending another reminder email, Active Nudge detects the bottleneck and proactively reaches out:

Active Nudge: Rahul joins in 3 days, but his onboarding is waiting for your input. What laptop and software access does he need?

The manager can simply respond:

Manager: MacBook please. He needs GitHub and Figma access. No Jira for now.

Active Nudge understands the response and extracts:

💻 MacBook → Required
GitHub → Required
Figma → Required
Jira → Not required

The agent then coordinates the next actions and updates the onboarding workflow.

The core loop is:

Detect → Understand → Act → Verify → Resolve

Unlike a traditional chatbot, Active Nudge doesn't wait for someone to ask a question. It identifies a bottleneck and proactively takes action.

How We Built It

We built an interactive frontend prototype representing the complete agentic workflow.

The prototype demonstrates:

A Freshservice-style onboarding workflow becoming stalled.
Active Nudge detecting the missing manager input.
The agent proactively initiating a conversation.
The manager responding naturally instead of filling another form.
AI extracting structured onboarding requirements.
Multiple specialized agents coordinating the workflow.
The onboarding workflow being updated and unblocked.

The architecture is designed around:

Orchestrator Agent → detects stalled onboarding and coordinates the workflow

Communication Agent → proactively communicates with the manager

Requirement Extraction Agent → converts natural-language responses into structured requirements

Provisioning / Update Agent → updates the onboarding workflow and triggers downstream actions

For the planned implementation, Freshworks Agent Studio acts as the orchestration layer, while MCP provides governed access to tools and external systems.

Our Stage 1 prototype focuses on demonstrating the complete product experience. The simulated actions are designed to be replaced with real Freshworks Agent Studio and MCP integrations during the 24-hour build.

Challenges We Ran Into

The biggest challenge was avoiding the obvious solution: another AI chatbot.

A chatbot could answer questions about onboarding, but it wouldn't solve the actual problem. The problem is that the workflow is stuck.

We therefore designed Active Nudge around autonomous action.

The agent needs to:

Detect that something is blocked.
Understand why it is blocked.
Identify who can unblock it.
Proactively communicate with them.
Understand an unstructured response.
Convert it into actionable information.
Execute the appropriate workflow.
Escalate when autonomous execution isn't possible.

We also had to balance our vision with the constraints of a hackathon prototype. Rather than spending our limited time building multiple enterprise integrations, we focused on making the complete agent experience and architecture clear and extensible.

Accomplishments We're Proud Of

We're proud that Active Nudge demonstrates agentic behavior rather than simply adding AI to an existing workflow.

The key transformation is:

Traditional approach

Workflow stuck → Send reminder → Wait → Send another reminder → Manual intervention

Active Nudge

Workflow stuck → Detect → Proactively communicate → Understand → Execute → Verify

We also designed the system around specialized agents instead of one monolithic AI assistant, allowing each agent to focus on a specific responsibility while an orchestrator coordinates the overall workflow.

Most importantly, we created a prototype where the entire story—from detecting the bottleneck to resolving it—can be understood in a short live demo.

What We Learned

We learned that the most valuable agentic systems aren't necessarily the ones with the most complicated AI.

They are the systems that identify a real operational bottleneck and can take meaningful action to remove it.

We also learned that human-centered agent design matters. Managers shouldn't have to learn another interface just because an organization wants to automate onboarding. The agent should meet people where they already work and understand how they naturally communicate.

Finally, we learned that autonomy needs boundaries. A good agent should know when it can act, when it needs clarification, and when it should escalate to a human.