Auto Web Task Agent

“A Nova-powered AI agent that navigates websites and completes tasks automatically.”

Comment

Inspiration

Many everyday tasks on the internet are repetitive and time-consuming. Whether it is applying for internships, filling forms, searching products, or booking services, users often spend a lot of time navigating websites and clicking through multiple pages.

As students and developers, we noticed how much time is wasted performing the same manual actions again and again. We wanted to explore whether AI agents could automate these workflows in the same way a human interacts with a website.

This idea inspired us to build Auto Web Task Agent, an intelligent assistant that can understand natural language instructions and automatically perform tasks on websites.

What it does

Auto Web Task Agent allows users to complete web tasks using simple commands.

Instead of manually navigating websites, a user can simply type something like:

"Apply for AI internships on LinkedIn."

The AI agent will then:

Understand the user's intent

Break the task into logical steps

Navigate the website automatically

Fill required forms

Complete the task

The system transforms natural language into automated browser actions, reducing repetitive manual work.

How we built it

Our system combines AI reasoning, automation, and a simple user interface.

Core AI Layer

We used Amazon Nova to power the reasoning and planning capabilities of the agent. The model interprets user instructions and converts them into structured actions.

Automation Layer

Using Amazon Nova Act, the agent interacts with web interfaces, performs clicks, fills forms, and navigates pages.

Backend

The backend manages the agent workflow, task planning, and communication between the AI model and the browser automation system.

Frontend

We built a lightweight dashboard where users can enter commands and observe the task execution steps in real time.

Storage

Files such as resumes or input data are stored using Amazon S3.

The overall pipeline looks like this:

User Command → AI Reasoning → Task Planning → Web Automation → Task Completion

Challenges we ran into

One of the biggest challenges was designing a reliable workflow where the AI could convert human instructions into precise browser actions.

Websites have different layouts and dynamic elements, which makes automation difficult. We had to carefully structure prompts and agent workflows to ensure consistent behavior.

Another challenge was balancing AI reasoning with deterministic automation, ensuring that the agent could adapt to different tasks while still executing reliable steps.

Accomplishments that we're proud of

Successfully built an AI-powered web automation agent that can convert natural language instructions into real web actions.

Integrated Amazon Nova reasoning capabilities to understand user intent and generate step-by-step workflows.

Implemented browser automation using Amazon Nova Act, allowing the agent to interact with websites like a real user.

Designed a simple and intuitive interface where users can give commands and watch tasks being executed in real time.

Built a working prototype within a short hackathon timeframe that demonstrates the real potential of AI agents for productivity.

What we learned

During the development of Auto Web Task Agent, we gained valuable insights into building AI-driven applications:

How agentic AI systems break down complex instructions into smaller executable steps.

Integrating foundation models like Amazon Nova with real-world automation workflows.

Designing prompts and agent logic to ensure reliable task execution.

Handling dynamic web interfaces and ensuring automation works across different websites.

Building scalable cloud-based AI solutions using AWS services.

This project helped us better understand how AI agents can move beyond chatbots and actually perform real-world actions.

What's next for Auto Web Task Agent

We plan to expand Auto Web Task Agent into a more powerful productivity platform. Future improvements include:

Supporting more complex multi-step workflows across multiple websites.

Adding voice commands so users can speak tasks naturally.

Building personalized agents that remember user preferences.

Integrating additional AI capabilities like multimodal understanding for documents and screenshots.

Expanding the system for use cases such as job applications, travel booking, research automation, and business workflows.

Our long-term vision is to create a fully autonomous digital assistant that can perform everyday online tasks with minimal human effort.

Built With

Share this project:

Updates