FlowPilot

FlowPilot Mission Control — The central dashboard for launching and monitoring autonomous AI workflows.
FlowPilot Mission History — Review completed missions, track their status, and access previous results and generated reports.
FlowPilot Workspace — Organize uploaded files, generated artifacts, datasets, and mission outputs in one centralized workspace.
Agent Memory — Persistent context that helps FlowPilot remember information and make smarter decisions across tasks.
Multi-Agent Fleet — AI agents divide tasks, share context, and coordinate execution to achieve goals autonomously.
Agent Settings — Configure agent behavior, capabilities, and execution preferences for each autonomous workflow.
New Mission — Define a goal and give FlowPilot the instructions needed to execute it autonomously.
Pilot AI — FlowPilot’s built-in AI that continuously understands the screen and assists users when issues arise.
Sample Final Report — A clear summary of the mission, completed tasks, results, and key outcomes.
Live Mission — Real-time view of FlowPilot executing tasks, coordinating agents, and tracking progress.
Live Mission — Real-time view of FlowPilot executing tasks, coordinating agents, and tracking progress.
Live Mission — Real-time view of FlowPilot executing tasks, coordinating agents, and tracking progress.

Inspiration

Most AI assistants still work like a chat. You ask something, get an answer, and then tell the AI what to do next.

I wanted to build something different: an AI where I can give it a goal, instead of giving it every single step.

That's why I built TaskPilot.

What it does

TaskPilot is an autonomous AI agent that takes a complex goal, breaks it into smaller tasks, works through those tasks, checks its progress, and handles problems when they occur.

The basic idea is:

Goal → Plan → Execute → Monitor → Recover → Complete

The user doesn't need to constantly tell the agent what to do next.

How I built it

I built TaskPilot using the technologies required by the challenge:

Gemini 3.5+ for AI reasoning and decision-making
Google ADK for the agent architecture
Cloud Run for backend execution
Pub/Sub for asynchronous tasks
Firestore for storing mission state and history
Cloud Storage for files and generated results

The application has an interface where users can create missions and watch the agent work through them.

For complex missions, the system can divide the work between different specialized agents, such as planning, analysis, data processing, validation, and recovery.

What I learned

While building TaskPilot, I learned that making an AI agent is much more than connecting a model to a chat interface.

The agent needs to keep track of its tasks, decide what should happen next, deal with failures, and know when the work is actually finished.

I also learned that autonomy and speed have to be balanced. An agent shouldn't perform unnecessary steps just because it can.

Challenges

One of the biggest challenges was designing the system so that it could handle complex workflows without becoming unnecessarily slow.

Another challenge was connecting the AI reasoning with cloud infrastructure. The AI decides what needs to happen, while the cloud services help run, store, and track the work.

What's next

I want to continue improving TaskPilot with better tool use, faster execution, more reliable recovery, and support for more types of complex workflows.

The main idea behind the project is simple:

Give the AI the goal, not every step.