Closer

CLOSER is an autonomous AI agent that finds unfinished tasks across your digital life, handles the repetitive work in the background, and only asks you when a real decision is needed.

Comment

Inspiration

Modern life is full of unfinished loops.

A refund needs a follow-up. A warranty claim needs documents. An appointment needs to be rescheduled. A company asks for information that is buried somewhere in your files. None of these tasks is individually difficult, but together they consume hours of human attention.

We asked a simple question:

What if an AI agent didn't just remind people about unfinished work, but actually finished it?

That idea became CLOSER.

What CLOSER Does

CLOSER is an autonomous AI agent for closing repetitive administrative loops.

Instead of acting like another productivity app that requires constant management, CLOSER works in the background. It discovers unfinished work, understands the situation, gathers relevant evidence, determines the next action, executes safe actions, verifies the outcome, and only interrupts the user when a meaningful human decision is required.

Its core loop is:

Discover → Understand → Gather → Decide → Act → Verify → Close

For example, when CLOSER discovers a potential warranty issue, it can find the purchase receipt, identify the product and purchase date, check warranty eligibility, gather the required evidence, prepare the claim, and track the result. If an external action requires human approval, CLOSER stops and presents the user with a clear decision instead of acting beyond its authority.

Why It Matters

Today's AI assistants are often reactive: the human asks, the AI responds.

CLOSER is designed to be proactive.

The goal is not to give people another interface to manage. The goal is to remove the management itself.

CLOSER treats human attention as the scarce resource. Routine work should happen quietly in the background. Human attention should only be requested when judgment, authorization, or an important decision is genuinely required.

How We Built It

CLOSER is built around the Strands Agents SDK, with a supervisor coordinating specialized agents for intake, evidence gathering, resolution planning, communication, and verification.

The system separates AI reasoning from execution authority. An agent can recommend an action, but a deterministic policy layer decides whether that action is actually permitted. Higher-risk actions can require explicit human approval.

The system also maintains structured state for every open loop, including its evidence, status, actions, approvals, and verification results.

Background execution allows CLOSER to continue working without requiring the user to constantly interact with it.

What Makes CLOSER Different

CLOSER is not another chatbot, reminder app, or task manager.

A reminder tells you that something needs to be done.

A task manager records what you need to do.

CLOSER tries to get the thing done.

It also knows when to stay silent.

A successful low-risk action should not generate another notification. A financial decision, sensitive action, or ambiguous situation should surface clearly for human approval.

This creates a different relationship between people and AI:

Don't manage the agent. Let the agent manage the work.

Challenges We Faced

Building an autonomous agent creates a difficult balance between autonomy and control.

Giving an AI agent too little authority makes it just another assistant that constantly asks questions. Giving it too much authority creates unacceptable risks.

We therefore designed CLOSER around explicit action risk levels, approval boundaries, evidence-backed decisions, idempotent actions, verification after execution, and an auditable state machine.

We also designed the demo around synthetic data so the entire workflow can be reproduced reliably while still exercising the real agent architecture.

What's Next

CLOSER can eventually connect to the services people already use: email, calendars, cloud documents, purchases, bills, service providers, and other everyday systems.

The long-term vision is simple:

People should not have to remember everything that needs to happen.

CLOSER should remember, act, follow up, verify, and only come back when the human is truly needed.

Built With

Share this project:

Updates

Submission history